JP6980231B2 - Communication status prediction device, communication status prediction method, and program - Google Patents

Description

The present invention relates to a technique for predicting a communication status and a communication status of wireless communication.

In a narrow space such as a factory, unexpected deterioration of communication quality (QoS: Quality of Service) occurs due to changes in the propagation environment due to the movement of machine tools, products, workers, etc., and interference between multiple wireless applications. Productivity is reduced.

Therefore, it is necessary to take evasive action before the extreme deterioration of QoS (communication failure) occurs, and it is necessary to predict the occurrence of communication failure in order to realize it. By predicting communication failures, for example, the following measures can be taken to prevent productivity decline.
(1) When a sign of communication failure is detected in the communication link of a certain wireless terminal, the transmission throughput can be reduced by stopping other wireless terminals using the same wireless channel or changing the bandwidth used. To prevent.
(2) When a sign of a communication failure is detected on a certain communication link, the bandwidth of the terminal causing the interference is automatically limited to prevent a decrease in transmission throughput and a large latency. ..

For example, Patent Documents 1 and 2 disclose a technique for predicting communication quality at a future time based on past information.

Japanese Unexamined Patent Publication No. 2016-91271 Japanese Unexamined Patent Publication No. 2017-5473

When the techniques of Patent Documents 1 and 2 are used for detecting a sign of a communication failure, a two-choice determination is made as to whether or not the predicted value satisfies the required QoS value, or the degree of similarity between the required QoS value and the predicted value (for example). , The Euclidean distance between the required QoS value and the predicted value) can be considered as a method of quantifying and determining the risk of communication failure. When making predictions based on past information in this way, it is inevitable that frequent input patterns can be predicted with high accuracy, but for rare input patterns or unlearned input patterns, the prediction accuracy deteriorates. Therefore, when the above-mentioned technology is used, it causes a problem that communication failure is not detected or erroneously detected.

Therefore, in view of the above problems, the present invention is a communication status prediction device and a communication status prediction device that execute highly accurate communication failure prediction processing even when a rare input pattern or an unlearned input pattern occurs. The purpose is to realize the method and the program.

In order to solve the above problems, the first invention is a communication status prediction device including a prediction value acquisition unit, a prediction error distribution acquisition unit, and a risk value acquisition unit.

The predicted value acquisition unit acquires the predicted value of the communication quality data after a predetermined time from the time series pattern of the communication quality data input at the current time.

The prediction error distribution acquisition unit acquires a prediction error distribution, which is a distribution of prediction error values, which is the difference between the predicted value of the communication quality data acquired by the predicted value acquisition unit and the true value of the predicted value.

The risk value acquisition unit acquires a risk value indicating the possibility of a communication failure at a future time based on the prediction error distribution and the predicted value.

This communication status prediction device acquires a risk value indicating the possibility of a communication failure at a future time based on the prediction error distribution and the predicted value, so that a communication failure occurs at a future time based on the acquired risk value. It is possible to properly grasp the possibility of occurrence. Further, in this communication status prediction device, the risk value can be acquired as continuous value or multi-value data by processing as described above, so that binary data indicating the presence or absence of a communication failure as in the prior art. Compared with the judgment processing by, it is possible to realize a much more accurate prediction processing.

The second invention is the first invention, further comprising a rarity acquisition unit.

The rarity acquisition unit acquires a rarity value indicating the degree of rarity of the occurrence frequency of the time series pattern of the communication quality data, which is the data indicating the communication quality.

The prediction error distribution acquisition unit determines a plurality of classes based on the rarity value, and acquires a prediction error distribution for each determined class.

The risk value acquisition unit acquires a risk value indicating the possibility of a communication failure at a future time based on the predicted value and the prediction error distribution corresponding to the class determined based on the rarity value.

In this communication status prediction device, the class to be classified is specified based on the rarity value of the time series data of the observation data (for example, QoS value) to be processed, and the prediction error distribution data corresponding to the class and the future time are specified. Based on the predicted value, a risk value indicating the probability of occurrence of a communication failure at the future time is acquired. Then, in this communication status prediction device, it is possible to appropriately grasp the possibility that a communication failure will occur at a future time based on the acquired risk value. Then, in this communication status prediction device, the risk value can be acquired as continuous value or multi-value data by processing as described above, so that binary data indicating the presence or absence of a communication failure as in the prior art. Compared with the judgment processing by, it is possible to realize a much more accurate prediction processing.

In this way, in this communication status prediction device, while considering the rarity of the input pattern (time-series data of observation data (for example, QoS value)), which is an important factor when detecting a communication failure. Since it is possible to acquire a risk value indicating the possibility of a communication failure, it is possible to realize highly accurate prediction processing of the occurrence of a communication failure.

Further, since this communication status prediction device executes a prediction process of the occurrence of a communication failure in consideration of the rarity of the input pattern (time-series data of the observation data (for example, QoS value)), it is a rare input. Even when a pattern or an unlearned input pattern occurs, a risk value indicating the possibility of a communication failure can be appropriately acquired, and as a result, highly accurate prediction processing of the communication failure can be realized.

The third invention is the second invention, and the learning unit learns using the time-series pattern generation pattern of the communication quality data and the communication quality data of the future time after the time-series pattern is generated. Execute the process.

The storage unit is a data set for rarity calculation including a time-series pattern generation pattern of communication quality data acquired by the learning unit, and communication quality data at a future time after a predetermined time after the time-series pattern is generated. Stores trained data including with a label dataset containing.

The rarity acquisition unit calculates the norms of the time-series pattern of the communication quality data acquired at the current time and the plurality of time-series patterns of the communication quality data included in the data set for rarity calculation, and out of the calculated norms. Get the minimum value of as a rarity value.

In this communication status prediction device, by learning the generation pattern of the time-series data of the observation data (for example, QoS value) at the time of learning, the generation pattern of the time-series data of the observation data (for example, QoS value) that frequently appears can be obtained. grasp. Further, in this communication status prediction device, the similarity between the generation pattern of the time-series data of the observation data (for example, QoS value) acquired at the time of learning and the time-series data of the actual observation data (for example, QoS value) is determined. For example, it is determined by calculating the norm (for example, the Euclidean distance), and the rarity value indicating the rarity of the appearance of the time series data of the observation data (for example, the QoS value) is calculated. In this communication status prediction device, highly accurate communication failure prediction processing can be executed using the rarity value acquired as described above.

The fourth invention is the second or third invention, in which the prediction error distribution acquisition unit acquires the rarity distribution data by acquiring the frequency distribution of the rarity values at a predetermined time, and the acquired rarity distribution. Acquires a threshold group for classifying the rarity values included in the above into a predetermined number of classes.

As a result, this communication status prediction device can acquire a rarity distribution, which is a distribution of rarity values, and classify the data group of rarity values into a predetermined class by a threshold value based on the rarity distribution.

The fifth invention is any one of the first to fourth inventions, further comprising a subtraction unit.

The subtraction unit acquires a prediction error value which is the difference between the predicted value of the communication quality data acquired by the predicted value acquisition unit and the true value of the predicted value.

The prediction error distribution acquisition unit classifies and classifies the prediction error values into a class determined based on the rarity value and the threshold group of the time series pattern of the communication quality data used when acquiring the prediction error value. The prediction error distribution data of the prediction error value is acquired for each class.

Further, when the center of the prediction error distribution data corresponding to the class determined by the rarity value of the time series pattern is set as the predicted value, the prediction error distribution acquisition unit integrates the region below the required reference value of the communication quality data. Get the value that was set as the risk value.

As a result, in this communication status prediction device, the risk (possibility) of falling below the required standard value of communication quality data can be appropriately grasped from the above integrated value, and as a result, a communication failure may occur due to the risk value. It is possible to appropriately grasp the sex (the possibility that a situation below the required standard value of communication quality data may occur). Therefore, this communication status prediction device grasps the possibility of communication failure based on the risk value, and is highly accurate even when a rare input pattern or an unlearned input pattern occurs. It is possible to realize the prediction processing of various communication failures.

The sixth invention is the fifth invention, in which the prediction error distribution acquisition unit determines the prediction error distribution data for each classified class based on the average value and the variance value of the prediction error values belonging to the class. Set to normal distribution.

As a result, this communication status prediction device can acquire the risk value by using the prediction error distribution data approximated by the normal distribution.

The seventh invention is a communication situation prediction device including a first learning prediction unit and a second learning prediction unit.

The first learning prediction unit includes a first learning unit, a first storage unit, and a first prediction determination processing unit.

The first learning unit executes the learning process using the time-series pattern generation pattern of the communication quality data and the communication quality data of the future time after the time-series pattern is generated.

The first storage unit is a data set for calculating rarity including a time-series pattern generation pattern of communication quality data acquired by the first learning unit, and a future time after a predetermined time after the time-series pattern is generated. Stores trained data including with a label dataset containing communication quality data of.

The first prediction determination processing unit is a functional unit that performs prediction determination processing for the occurrence of a communication failure, and includes a first prediction value acquisition unit, a first prediction error distribution acquisition unit, and a first risk value acquisition unit. Be prepared.

The first predicted value acquisition unit acquires the predicted value of the communication quality data after a predetermined time from the time series pattern of the communication quality data input at the current time.

The first prediction error distribution acquisition unit acquires a prediction error distribution, which is a distribution of prediction error values, which is the difference between the predicted value of the communication quality data acquired by the predicted value acquisition unit and the true value of the predicted value.

The first risk value acquisition unit acquires a risk value indicating the possibility of a communication failure at a future time based on the prediction error distribution and the predicted value.

The second learning prediction unit includes a second learning unit, a second storage unit, and a second prediction determination processing unit.

The second learning unit executes the learning process using the time-series pattern generation pattern of the communication quality data and the communication quality data of the future time after the time-series pattern is generated.

The second storage unit is a data set for calculating rarity including a time-series pattern generation pattern of communication quality data acquired by the second learning unit, and a future time after a predetermined time after the time-series pattern is generated. Stores trained data including with a label dataset containing communication quality data of.

The second prediction determination processing unit is a functional unit that performs prediction determination processing for the occurrence of a communication failure, and includes a second prediction value acquisition unit, a second prediction error distribution acquisition unit, and a second risk value acquisition unit. ..

The second predicted value acquisition unit acquires the predicted value of the communication quality data after a predetermined time from the time series pattern of the communication quality data input at the current time.

The second prediction error distribution acquisition unit acquires a prediction error distribution, which is a distribution of prediction error values, which is the difference between the predicted value of the communication quality data acquired by the predicted value acquisition unit and the true value of the predicted value.

The second risk value acquisition unit acquires a risk value indicating the possibility of a communication failure at a future time based on the prediction error distribution and the predicted value.

And the first learning prediction unit
(1) When the second learning prediction unit is executing the prediction judgment process, the learning process is executed and the learning process is executed.
(2) When the second learning prediction unit is executing the learning process, the prediction determination process is executed.

As a result, in this communication status prediction device, the learning process and the prediction determination process can be executed by the pipeline process. As a result, since the communication status prediction device can always execute the learning process and the prediction determination process, it is possible to predict the communication failure with higher accuracy.

The eighth invention is a communication situation prediction method including a prediction value acquisition step, a prediction error distribution acquisition step, and a risk value acquisition step.

The predicted value acquisition step acquires the predicted value of the communication quality data after a predetermined time from the time series pattern of the communication quality data input at the current time.

The prediction error distribution acquisition step acquires a prediction error distribution, which is a distribution of prediction error values, which is the difference between the predicted value of the communication quality data acquired by the predicted value acquisition step and the true value of the predicted value.

The risk value acquisition step acquires a risk value indicating the possibility of a communication failure at a future time based on the prediction error distribution and the predicted value.

As a result, it is possible to realize a communication status prediction method having the same effect as that of the first invention.

The ninth invention is a program for causing a computer to execute the communication situation prediction method according to the eighth invention.

This makes it possible to realize a program for causing a computer to execute a communication status prediction method having the same effect as that of the first invention.

According to the present invention, a communication status prediction device, a communication status prediction method, and a communication status prediction device that execute highly accurate communication failure prediction processing even when a rare input pattern or an unlearned input pattern occurs. The program can be realized.

The schematic block diagram of the wireless condition prediction apparatus 100 (an example of a communication condition prediction apparatus) which concerns on 1st Embodiment. The schematic block diagram of the prediction determination processing unit 2 of the radio condition prediction apparatus 100 which concerns on 1st Embodiment. The figure which shows the time-series data (example) of the QoS value acquired by a wireless condition prediction apparatus 100. The figure for demonstrating the process of the 1st step executed in the learning period in a wireless condition prediction apparatus 100. The figure for demonstrating the process of the 2nd step executed in the learning period in the radio condition prediction apparatus 100. The figure for demonstrating the process of the 3rd step executed in the learning period in the wireless condition prediction apparatus 100. The figure for demonstrating the prediction processing executed in the prediction processing execution period in a wireless condition prediction apparatus 100. The figure for demonstrating the prediction processing (the prediction execution processing of a communication failure). The schematic block diagram of the wireless condition prediction apparatus 200 of 2nd Embodiment. A timing chart showing time-series data (example) of QoS values acquired by the wireless situation prediction device 200, a processing mode of the first learning prediction unit Dev1, and a processing mode of the second learning prediction unit Dev2. The figure which shows the CPU bus configuration.

[First Embodiment]
The first embodiment will be described below with reference to the drawings.

<1.1: Configuration of wireless status prediction device>
FIG. 1 is a schematic configuration diagram of a radio condition prediction device 100 (an example of a communication condition prediction device) according to the first embodiment.

FIG. 2 is a schematic configuration diagram of a prediction determination processing unit 2 of the radio condition prediction device 100 according to the first embodiment.

As shown in FIG. 1, the wireless status prediction device 100 predicts the antenna Ant1, the demodulation processing unit Dm1, the observation data acquisition unit FE1, the control unit CPU1, the learning unit 1, the learned data storage unit St1. A determination processing unit 2 is provided.

The antenna Ant1 is an antenna for receiving an RF signal (for example, a carrier band OFDM signal) from the outside. The antenna Ant1 outputs the received RF signal (for example, a carrier band OFDM signal) as a signal RFin to the demodulation processing unit Dm1. The antenna Ant1 may be an antenna (transmission / reception antenna) that also functions as a transmission antenna.

The demodulation processing unit Dm1 includes an RF processing unit Dm11 and a BB processing unit Dm12.

The RF processing unit Dm11 executes RF demodulation processing on the signal RFin and acquires a baseband signal.

The BB processing unit Dm12 executes baseband demodulation processing on the baseband signal demodulated by the RF processing unit Dm11, acquires the signal (data) after the baseband demodulation processing as data D0, and acquires the acquired data D0. Is output to the observation data acquisition unit FE1.

The observation data acquisition unit FE1 inputs the data D0 output from the demodulation processing unit Dm1 and acquires the QoS value which is the data indicating the QoS of the wireless communication environment in which the wireless condition prediction device 100 is placed from the data D0. do. Then, the observation data acquisition unit FE1 outputs the acquired data including the QoS value as data Din to the learning unit 1 and the prediction determination processing unit 2.

The QoS value is an index value indicating communication quality such as throughput, number of (continuous) packet losses, and delay amount.

The control unit CPU 1 is a functional unit for controlling each functional unit of the wireless status prediction device 100. The control unit CPU 1 outputs a mode signal Mode including data indicating the processing mode (for example, (1) learning mode, (2) prediction processing execution mode) of the wireless status prediction device 100 to the learning unit 1 and the prediction determination processing unit 2. do. Further, the control unit CPU 1 outputs the control signal ctl1 to the prediction determination processing unit 2.

The learning unit 1 inputs the data Din output from the observation data acquisition unit FE1. The learning unit 1 executes a learning process for executing a pattern matching system prediction process by PNN (Probabilistic Nuclear Network) or the like on the data Din, that is, the time series data of the QoS value, and acquires the learned data. .. As the trained data, for example, (1) a data set for Rarity calculation and (2) a data set for future value prediction are acquired (details will be described later). The learning unit 1 stores the acquired learned data as data D_trained in the learned data storage unit St1.

The learned data storage unit St1 stores the learned data output from the learning unit 1. Further, the learned data storage unit St1 reads data according to a request from the prediction determination processing unit 2, and outputs the read data (for example, data D_trained) to the prediction determination processing unit 2.

As shown in FIG. 2, the prediction determination processing unit 2 includes a data holding unit 21, a rarity acquisition unit 22, a prediction value acquisition unit 23, a subtractor 24, a prediction error distribution acquisition unit 25, and a risk value acquisition unit. 26 and.

The data holding unit 21 inputs the data Din from the observation data acquisition unit FE1. The data holding unit 21 stores and holds the input data Din for a predetermined period. The data holding unit 21 outputs the time-series data of the data Din as the data D1s to the rarity acquisition unit 22 and the predicted value acquisition unit 23. When data Din at time i and is denoted as _{D i,} the data D1s, for example, time-series data _{D1s = {D i, D i} + 1, D i + 2} three consecutive set as. The number of data in the time-series data D1s may be any number other than 3 (k1 (k1: natural number) continuous time-series data (D1s = {D _i , _{Di + 1} , ...,). _{Di + k1-1} })).

Further, assuming that the predicted time is time i + k (k: natural number), the data holding unit 21 outputs the actual data Din (= _{Di + k} ) (true value) at time i + k to the subtractor 24 as data D1_true.

As shown in FIG. 2, the rarity acquisition unit 22 includes a norm acquisition unit 221 and a Rarity calculation unit 222.

The norm acquisition unit 221 inputs the trained data D_trained output from the trained data storage unit St1 and the data D1s output from the data holding unit 21. The norm acquisition unit 221 calculates the norm (for example, the Euclidean distance) using the trained data D_trained and the data D1s, and outputs the data including the calculated norm as the data D_norm to the Rarity calculation unit 222. When there are n1 (n1: natural numbers) calculated norms, the data D_norm is the data of a set including _{n1 norms (d 0} , d ₁ , ..., d _n1-1). At this time, the data D_norm is
D_norm = {d ₀ , d ₁ , ..., d _n1-1 }
Notated as.

The Rarity calculation unit 222 inputs the data D_norm output from the norm acquisition unit 221 and calculates the rarity indicating the rarity of the appearance of the time series data D1s from the data D_norm. Then, the Rarity calculation unit 222 outputs the data including the calculated rarity as the data D2 to the prediction error distribution acquisition unit 25.

The predicted value acquisition unit 23 inputs the trained data D_trained output from the trained data storage unit St1 and the data D1s output from the data holding unit 21. The predicted value acquisition unit 23 performs prediction processing (for example, prediction processing using PNN) using the trained data D_trained and data D1s, and acquires the value of the prediction data at a future time as data D_predict. Then, the predicted value acquisition unit 23 outputs the acquired data D_predict to the subtractor 24 and the risk value acquisition unit 26.

The subtractor 24 inputs the data D1_true output from the data holding unit 21 and the data D_predict output from the predicted value acquisition unit 23. The subtractor 24
D_prediction_error = D1_tree-D_predict
The prediction error data D_prescription_error is acquired by executing the process (subtraction process) corresponding to. Then, the subtractor 24 outputs the acquired prediction error data D_prescription_error to the prediction error distribution acquisition unit 25.

As shown in FIG. 2, the prediction error distribution acquisition unit 25 includes a selector SEL1, a Rality distribution acquisition unit 251, a classification processing unit 252, a first storage unit St2, a Rarity class classification unit 253, and a prediction error distribution. It includes a data acquisition unit 254, a second storage unit St3, and a class determination unit 255.

The selector SEL1 is a selector with 1 input and 3 outputs, and inputs the data D2 output from the Rarity calculation unit 222 of the rarity acquisition unit 22 and the control signal ctl1 output from the control unit CPU1. The selector SEL1 outputs the input data D2 to any one of (1) Rarity distribution acquisition unit 251, (2) Rarity class classification unit 253, and (3) class determination unit 255 according to the control signal ctl1.

The Rarity distribution acquisition unit 251 inputs the data D2 output from the selector SEL1. The Rarity distribution acquisition unit 251 acquires the Rarity distribution from the data D2, and outputs the acquired Rarity distribution data as data Dist_Rarity to the classification processing unit 252.

The class classification processing unit 252 inputs the data Dist_Rarity output from the Rarity distribution acquisition unit 251. The class classification processing unit 252 executes the class classification process based on the data Dist_Rarity, acquires the threshold set data Th for class classification, and stores the acquired threshold set data Th in the first storage unit St2. ..

When the thresholds required for classifying are n2 (n2: natural numbers), the set data Th of the thresholds has n2 thresholds (th ₀ , th ₁ , ..., Th _n2-1 ). It is the data of the set including. At this time, the set data Th of the threshold value is
Th = {th ₀ , th ₁ , ..., th _n2-1 }
Notated as.

The first storage unit St2 stores the set data Th of the threshold values output from the class classification processing unit 252. Further, the first storage unit St2 reads the data according to the data read request from the Rarity class classification unit 253, and outputs the data to the Rarity class classification unit 253. Further, the first storage unit St2 reads data according to the data read request from the class determination unit 255, and outputs the data to the class determination unit 255.

The Rarity class classification unit 253 inputs the data D2 output from the selector SEL1 and the set data Th of the threshold values output from the first storage unit St2. The Rarity class classification unit 253 classifies the data D2 based on the set data Th of the threshold value, and outputs the data including the class number j to which the data D2 is distributed as the data D3 to the prediction error distribution data acquisition unit 254. do.

The prediction error distribution data acquisition unit 254 inputs the data D_output_error output from the subtractor 24 and the data D3 output from the Rarity class classification unit 253. The prediction error distribution data acquisition unit 254 distributes the data D_prescription_error to classes based on the data D3, and acquires the prediction error distribution data by acquiring the occurrence frequency distribution of the data D_prescription_error for each distributed class. Then, the prediction error distribution data acquisition unit 254 stores the acquired data _{including the prediction error distribution data Err_dist j} as data D4 in the second storage unit St3. The prediction error distribution data of the prediction error classified into the class of the class number _j is referred to as Err_dist j.

The second storage unit St3 stores the data D4 (= Err_dist _j ) output from the prediction error distribution data acquisition unit 254. Further, the second storage unit St3 reads data in accordance with the data read request from the risk value acquisition unit 26, and outputs the data to the risk value acquisition unit 26.

The class determination unit 255 inputs the data D2 output from the selector SEL1 and the set data Th of the threshold values output from the first storage unit St2. The class determination unit 255 classifies the data D2 based on the set data Th of the thresholds, and class number j (= class) which is information for specifying the class to which the data D2 is distributed (for example, the class whose class number is j). (R _i ), class (x) is a function for acquiring the class number of x), and outputs the data including the data to the risk value acquisition unit 26.

The risk value acquisition unit 26 includes data D_predicted output from the predicted value acquisition unit 23, class number j output from the class determination unit 255, and data D4 (= Err_dist _j ) output from the second storage unit St3. , The data QoS_req indicating the required reference value of the communication quality data and the mode signal Mode output from the control unit CPU 1 are input. When the mode signal Mode is a signal indicating the "prediction processing execution mode", the risk value acquisition unit 26 is the time when the true value of the data D_predicted is acquired based on _{the prediction error distribution data D4 (= Err_dist j) (future).} At time), acquire the risk value indicating the possibility of communication failure. Then, the risk value acquisition unit 26 outputs the data including the acquired risk value as data Dout.

The data QoS_req indicating the required reference value of the communication quality data is preset in the data set in the wireless status prediction device 100 (for example, stored in a storage unit (not shown) in the wireless status prediction device 100). It may be data) or data set from the outside of the wireless condition prediction device 100.

<1.2: Operation of wireless status prediction device>
The operation of the radio condition prediction device 100 configured as described above will be described below.

FIG. 3 is a diagram showing time-series data (example) of QoS values acquired by the radio condition prediction device 100. In the graph of FIG. 3, the horizontal axis is time and the vertical axis is a QoS value.

FIG. 4 is a diagram for explaining the processing of the first step executed in the learning period in the radio condition prediction device 100.

FIG. 5 is a diagram for explaining the process of the second step executed in the learning period in the radio condition prediction device 100.

FIG. 6 is a diagram for explaining the processing of the third step executed in the learning period in the radio condition prediction device 100.

FIG. 7 is a diagram for explaining the prediction process executed in the prediction process execution period in the wireless condition prediction device 100.

(1.2.1: Learning process)
First, the learning process will be described. The learning process is divided into (1) a process of the first step, (2) a process of the second step, and (3) a process of the third step, and is executed.
(1) In the processing of the first step, the learning vector and label acquisition processing is executed from the time series data (observed value series) of the QoS value.
(2) In the process of the second step, the threshold value for class determination is calculated based on the Rarity distribution.
(3) In the process of the third step, data for calculating the prediction error distribution of the QoS value is acquired.

Hereinafter, the following will be described separately for (1) first step processing, (2) second step processing, and (3) third step processing.

(1.2.1.1: Processing of the first step)
In the process of the first step, the radio condition prediction device 100 executes the acquisition process of the learning vector and the label from the time series data (observed value series) of the QoS value.

The demodulation processing unit Dm1 of the radio condition prediction device 100 executes demodulation processing on the radio signal received via the antenna Ant1 and acquires an RF demodulation signal and a BB demodulation signal. The observation data acquisition unit FE1 is an index indicating communication quality such as a QoS value (throughput, (continuous) packet loss number, delay amount, etc.) from the RF demodulation signal and / or the BB demodulation signal acquired by the demodulation processing unit Dm1. Value). Then, the observation data acquisition unit FE1 outputs the acquired data including the QoS value as data Din to the learning unit 1.

The control unit CPU 1 sets the mode signal Mode to a signal value indicating the "learning mode" and outputs the mode signal to the learning unit 1.

The learning unit 1 executes a learning process for executing a pattern matching system prediction process by PNN (Probabilistic Nuclear Network) or the like on the data Din, that is, the time series data of the QoS value, and acquires the learned data. .. Specifically, the learning unit 1 processes as follows (1) to (3).
(1) Acquire k1 time-series observation value series (QoS value time-series data) from time i (corresponding to time t0 in the case of FIG. 4), and acquire a learning vector. Further, the future value data (correct answer data) corresponding to the learning vector is set as the observed value (QoS value) at time i + k2 on the label associated with the learning vector. For example, in the case of FIG. 4, i = 0, k1 = 3, and the time series data D1s of the QoS value is
D1s = {D _i , _{Di + 1} , _{Di + 2} }
= {A _i , a _{i + 1} , a _{i + 2} } = {a ₀ , a ₁ , a ₂ }
The above data D1s is used as a learning vector. Then, in the case of FIG. 4, since k2 = 4, the label associated with the learning vector is the observed value (QoS value) _{ai + 4} at the time i + k2 (= i + 4). The target data of the above processing in the case of FIG. 4 is the data included in the section of the window Win1 of FIG. 4 (the section of the time t0 to tw).
(2) Next, advance the time (time step) by one. That is, the section of the window Win1 in FIG. 4 is shifted to Dir1 in the time axis direction by one hour step (time corresponding to one data sampling), and the shifted section of the window Win1 is used as the target data for processing.

Similar to the process (1), k1 time-series observed value series (time-series data of QoS values) are acquired from time i + 1, and a learning vector is acquired. Further, the future value data (correct answer data) corresponding to the learning vector is set as the observed value (QoS value) at time i + k2 + 1 on the label associated with the learning vector. For example, in the case of FIG. 4, i = 0, k1 = 3, and the time series data D1s of the QoS value is
D1s = {Di _{+ 1} , _{Di + 2} , _{Di + 3} }
= {A _{i + 1} , a _{i + 2} , a _{i + 3} } = {a ₁ , a ₂ , a ₃ }
The above data D1s is used as a learning vector. Then, in the case of FIG. 4, since k2 = 4, the label associated with the learning vector is the observed value (QoS value) _{ai + 4} at the time i + k2 (= i + 4).
(3) The learning unit 1 repeatedly executes the processes (1) and (2) above.

At time i + n1-1 (n1: natural number), as in processing (1), k1 time-series observed value series (time-series data of QoS values) are acquired from time i + n1-1, and a learning vector is acquired. do. Further, the future value data (correct answer data) corresponding to the learning vector is set as the observed value (QoS value) at time i + n1-1 + k2 on the label associated with the learning vector. For example, in the case of FIG. 4, i = 0, k1 = 3, and the time series data D1s of the QoS value is
D1s = {D _{i + n1-1} , _{Di + n1} , _{Di + n1 + 1} }
_{= {A i + n1-1, a} i + n1, a i + n1 + 1} = {a n1-1, a n1, a n1 + 1}
The above data D1s is used as a learning vector. In the case of FIG. 4, since it is k2 = 4, the label to be associated with the learning vector, observed value at time i + n1-1 + k2 (= i + n1 + 3) (QoS _values) and _{a n1 + 3.}

From the data acquired as described above, the learning unit 1 acquires the future value prediction data set Dset1, the Rarity calculation data set Dset2, and the label (correct answer data) data set Dset_Label.

For convenience of explanation, the number of label data corresponding to k1 = 3, k2 = 2, and one learning vector is set to one (corresponding to the case shown in FIG. 4), and the following description will be made. Further, the number of label data corresponding to one learning vector of k1, k2 is not limited to the above number.

なお、上記のＲａｒｉｔｙ計算用データセットＤｓｅｔ２の行列において、各行ベクトルが、時刻ｉの学習ベクトルに対応する。 In the above case (corresponding to the case shown in FIG. 4), the future value prediction data set Dset1, the Rarity calculation data set Dset2, and the label (correct answer data) data set Dset_Label are as follows by a matrix. Can be expressed.

In the matrix of the Rarity calculation data set Dset2, each row vector corresponds to the learning vector at time i.

Using the above data set, the learning unit 1 executes, for example, a learning process by PNN, and acquires a trained model (for example, a learning model by PNN). As the processing by PNN, for example, the techniques disclosed in Japanese Patent Application No. 2017-232442 and Japanese Patent Application No. 2018-042747 may be used.

Then, the learning unit 1 is a data set of the data acquired as described above ((1) future value prediction data set Dset1, (2) Rarity calculation data set Dset2, and (3) label (correct answer data) data set. Dset_Label) is stored in the trained data storage unit St1 as the trained data D_trained.

(1.2.1.2: Processing of the second step)
In the process of the second step, the radio condition prediction device 100 acquires a Rarity distribution from the time series data (observed value series) of the QoS value, and calculates a threshold value for class determination based on the Rarity distribution. Run.

The demodulation processing unit Dm1 of the radio condition prediction device 100 executes demodulation processing on the radio signal received via the antenna Ant1 and acquires an RF demodulation signal and a BB demodulation signal. The observation data acquisition unit FE1 acquires a QoS value from the RF demodulation signal and / or the BB demodulation signal acquired by the demodulation processing unit Dm1. Then, the observation data acquisition unit FE1 outputs the data including the acquired QoS value as the data Din to the data holding unit 21 of the prediction determination processing unit 2.

The control unit CPU 1 sets the mode signal Mode to a signal value indicating the “learning mode” and outputs the mode signal to the prediction determination processing unit 2. Further, the control unit CPU1 sets the control signal ctl1 to a signal value (a signal value at which the terminal 0 of the selector SEL1 is selected) for executing the "processing of the second step", and outputs the control signal ctl1 to the prediction determination processing unit 2. ..

The data holding unit 21 stores and holds the data Din from the observation data acquisition unit FE1 for a predetermined period. The data holding unit 21 outputs the time-series data of the data Din as the data D1s to the rarity acquisition unit 22 and the predicted value acquisition unit 23. When data Din at time i and is denoted as _{D i,} the data D1s, for example, time-series data _{D1s = {D i, D i} + 1, D i + 2} three consecutive set as. For convenience of explanation, as shown in FIG. 5, the observed value series (time series data) of the QoS values after the time t1 is b ₀ , b ₁ , b ₂ , .... In this case, the data holding unit 21 outputs the data D1s = {b ₀ , b ₁ , b ₂ } to the rarity acquisition unit 22.

The norm acquisition unit 221 of the rarity acquisition unit 22 reads the learned data D_trained Rarity calculation data set Dset2 from the learned data storage unit St1 and inputs the data D1s output from the data holding unit 21. Here, a case where the norm acquisition unit 221 processes the data D1s = {b ₀ , b ₁ , b ₂ } will be described.

The norm acquisition unit 221 calculates the norm (for example, Euclidean distance) by using each row vector of the data set Dset2 for Rarity calculation and the data D1s (= {b ₀ , b ₁ , b _2}). In other words, the norm acquiring unit 221, when the m-th row of the row vectors of Rarity calculation data set Dset2 and Dset2_r _m,
d ₀ = Norm (Dset2_r ₀ , D1s)
d ₁ = Norm (Dset2_r ₁ , D1s)
d ₂ = Norm (Dset2_r ₂ , D1s)
・ ・ ・
d _n1-1 = Norm (Dset2_r _n1-1 , D1s)
Norm (v1, v2): A function to acquire the norms (for example, Euclidean distance) of the vectors v1 and v2.
The set data of the norm {d ₀ , d ₁ , ..., d _n1-1 } is acquired by the process corresponding to. Then, the norm acquisition unit 221 outputs the data including the acquired set data of the norms {d ₀ , d ₁ , ..., d _n1-1 } as the data D_norm to the Rarity calculation unit 222.

The Rarity calculation unit 222 calculates the rarity indicating the rarity of the appearance of the time series data D1s from the data D_norm output from the norm acquisition unit 221. Specifically, the Rarity calculation unit 222
_{r i = min (d i,} d i + 1, ···, d i + n1-1)
min (): By performing the process corresponding to the function to obtain the minimum value of the element, and acquires the time series data _{{D i, D i + 1} , D i + 2} rarity value _{r i} corresponding to the QoS value. Here, since the time series data of the QoS value {D _i , _{Di + 1} , _{Di + 2} } = {b ₀ , b ₁ , b ₂ }, the Rarity calculation unit 222 is used.
r ₀ = min (d ₀ , d ₁ , ..., d _n1-1 )
To get.

Then, Rarity calculating unit 222, the data including the calculated rarity _{r i} as data D2, and outputs to the selector SEL1 of the prediction error distribution obtaining unit 25.

The selector SEL1 selects and inputs the terminal 0 according to the control signal ctl1 set in the signal value for executing the "processing of the second step" (the signal value at which the terminal 0 of the selector SEL1 is selected). The data D2 is output to the Signal distribution acquisition unit 251.

The Rarity distribution acquisition unit 251 acquires the Rarity distribution from the data D2 output from the selector SEL1. Specifically, Rarity distribution obtaining unit 251, as shown in FIG. 5, each time the obtained rarity value r _i acquired predetermined period in (each sampling time), the frequency distribution diagram of rarity value ri ( Histogram) (corresponding to the lower right figure of FIG. 5) is acquired. The Rarity distribution acquisition unit 251 outputs the data for specifying the frequency distribution map (histogram) of the rarity value ri acquired as described above to the classification processing unit 252 as the Rarity distribution data Dist_Rarity.

The class classification processing unit 252 executes the class classification processing based on the data Dist_Rarity output from the Rarity distribution acquisition unit 251. Specifically, a clustering method such as the K-means method or a method by SOM (Self-organizing map) is applied to classify Rality into a predetermined number of classes, and the threshold value Th (= {th) for classifying the classes is applied. ₀ , th ₁ , ..., th _n2-1 }) is acquired.

Here, a case where the K-means method is adopted will be described.

Classification processing unit 252, based on the Rarity distribution data Dist_Rarity, the rarity value _{r i} a preset number of classes (number of clusters) KCl: when classifying into (KCl 2 or more natural number), by K-means clustering method , each rarity value r _i, the norm of the cluster center (e.g., Euclidean distance) is calculated, and the respective rarity value r _i, assigned to the nearest cluster center cluster. The class classification processing unit 252 repeatedly executes this processing, and if the status of cluster allocation does not change or the amount of change becomes equal to or less than a predetermined value, it is determined that the class has converged, and the class classification process is terminated. _{Then, the class classification processing unit 252 sets a threshold value Th (= {th 0} , th ₁ , ..., th _n2-1 }) for separating each class (each cluster) at the stage when the class classification process is completed. To set. It is assumed that the number of classes (number of clusters) kcl is preset in the class classification processing unit 252.

The class classification processing unit 252 transfers the set data Th (= {th ₀ , th ₁ , ..., th _{n2-1}) of the threshold values} for class classification acquired as described above to the first storage unit St2. Remember.

In the following, for convenience of explanation, as shown in the lower right figure of FIG. 5, the number of classes is set to "3", each class is set to class Class_0, class Class_1, and class Class_1, and the threshold value Th = {th ₀ , The _{case of setting th 1} } will be described.

(1.2.1.3: Processing in the third step)
In the process of the third step, the radio condition prediction device 100 acquires data for calculating the prediction error distribution of the QoS value.

The demodulation processing unit Dm1 of the radio condition prediction device 100 executes demodulation processing on the radio signal received via the antenna Ant1 and acquires an RF demodulation signal and a BB demodulation signal. The observation data acquisition unit FE1 acquires a QoS value from the RF demodulation signal and / or the BB demodulation signal acquired by the demodulation processing unit Dm1. Then, the observation data acquisition unit FE1 outputs the data including the acquired QoS value as the data Din to the data holding unit 21 of the prediction determination processing unit 2.

The control unit CPU 1 sets the mode signal Mode to a signal value indicating the “learning mode” and outputs the mode signal to the prediction determination processing unit 2. Further, the control unit CPU 1 sets the control signal ctl1 to a signal value (a signal value at which the terminal 1 of the selector SEL1 is selected) for executing the "processing of the third step", and outputs the control signal ctl1 to the prediction determination processing unit 2. ..

The data holding unit 21 stores and holds the data Din from the observation data acquisition unit FE1 for a predetermined period. The data holding unit 21 outputs the time-series data of the data Din as the data D1s to the rarity acquisition unit 22 and the predicted value acquisition unit 23. When data Din at time i and is denoted as _{D i,} the data D1s, for example, time-series data _{D1s = {D i, D i} + 1, D i + 2} three consecutive set as. For convenience of explanation, as shown in FIG. 6, the observed value series (time series data) of the QoS values after the time t2 is c ₀ , c ₁ , c ₂ , .... In this case, the data holding unit 21 outputs the data D1s = {c ₀ , c ₁ , c ₂ } to the rarity acquisition unit 22 and the predicted value acquisition unit 23.

_{Further, the data holding unit 21 outputs the data D1_true (= Di + k (c 4 in} the case of FIG. 6)) of the actual data (true value) of the time predicted by the predicted value acquisition unit 23 to the subtractor 24. do.

The norm acquisition unit 221 of the rarity acquisition unit 22 reads the learned data D_trained Rarity calculation data set Dset2 from the learned data storage unit St1 and inputs the data D1s output from the data holding unit 21. Here, a case where the norm acquisition unit 221 processes the data D1s = {c ₀ , c ₁ , c ₂ } will be described.

The norm acquisition unit 221 calculates the norm (for example, Euclidean distance) by using each row vector of the data set Dset2 for Rarity calculation and the data D1s (= {c ₀ , c ₁ , c _2}). In other words, the norm acquiring unit 221, when the m-th row of the row vectors of Rarity calculation data set Dset2 and Dset2_r _m,
d ₀ = Norm (Dset2_r ₀ , D1s)
d ₁ = Norm (Dset2_r ₁ , D1s)
d ₂ = Norm (Dset2_r ₂ , D1s)
・ ・ ・
d _n1-1 = Norm (Dset2_r _n1-1 , D1s)
Norm (v1, v2): A function to acquire the norms (for example, Euclidean distance) of the vectors v1 and v2.
The set data of the norm {d ₀ , d ₁ , ..., d _n1-1 } is acquired by the process corresponding to. Then, the norm acquisition unit 221 outputs the data including the acquired set data of the norms {d ₀ , d ₁ , ..., d _n1-1 } as the data D_norm to the Rarity calculation unit 222.

The Rarity calculation unit 222 calculates the rarity indicating the rarity of the appearance of the time series data D1s from the data D_norm output from the norm acquisition unit 221. Specifically, the Rarity calculation unit 222
_{r i = min (d i,} d i + 1, ···, d i + n1-1)
min (): By performing the process corresponding to the function to obtain the minimum value of the element, and acquires the time series data _{{D i, D i + 1} , D i + 2} rarity value _{r i} corresponding to the QoS value. Here, since the time series data of the QoS value {D _i , _{Di + 1} , _{Di + 2} } = {c ₀ , c ₁ , c ₂ }, the Rarity calculation unit 222 is used.
r ₀ = min (d ₀ , d ₁ , ..., d _n1-1 )
To get.

Then, Rarity calculating unit 222, the data including the calculated rarity _{r i} as data D2, and outputs to the selector SEL1 of the prediction error distribution obtaining unit 25.

The selector SEL1 selects and inputs the terminal 1 according to the control signal ctl1 set in the signal value (the signal value at which the terminal 1 of the selector SEL1 is selected) for executing the "processing of the third step". The data D2 is output to the Signaly class classification unit 253.

The Rarity class classification unit 253 reads the threshold set data Th (= {th ₀ , th ₁ }) from the first storage unit St2, and based on the threshold set data Th (= {th ₀ , th _1}), The data D2 is classified into classes, and the data including the class number j to which the data D2 is distributed is output as the data D3 to the prediction error distribution data acquisition unit 254.

The predicted value acquisition unit 23 reads the trained data D_trained (Rarity calculation data set Dset2 and label (correct answer data) data set Dset_Label) from the trained data storage unit St1. The predicted value acquisition unit 23 has the same time-series data D1s (= {c ₀ , c ₁ , c ₂ }) of the QoS value and each row vector of the data set Dset2 for Rarity calculation, as in the process executed by the norm acquisition unit 221. The norm (for example, Euclidean distance) is calculated, the row vector having the minimum value is detected, and the label (correct answer data) corresponding to the detected row vector is specified by using the data set Dset_Label. As a result, the predicted value acquisition unit 23 acquires the label corresponding to the row vector that takes the minimum value of the norm as the predicted data value D_predict of the future time. For example, in the case shown in FIG. 6, the time-series data D1s (= {c ₀ , c ₁ , c _{2 }) of the QoS value and the vector ({a 2} , a ₃ ) in the third row of the data set Dset2 for Rarity calculation. , A ₄ })), the predicted value acquisition unit 23 corresponds to the vector ({a ₂ , a ₃ , a ₄ }) in the third row of the data set Dset2 for Rarity calculation. specified as being predictive data value D_predict value _{a 6} of the label (label (data set first in the third row data of Dset_Label correct answer data)) of future time (future time i + 4 When the current time i) to.

_{Then, the predicted value acquisition unit 23 outputs the predicted data value D_predict (= D'i + k} = a ₆ ) of the future time (future time i + 4 when the current time i is assumed) acquired as described above to the subtractor 24.

The subtractor 24 inputs the data D1_true (= Di _{+ k} = c _{4 ) output from the data holding unit 21 and the data D_predict (= D'i + k} = a ₆ ) output from the predicted value acquisition unit 23. The subtractor 24
D_prediction_error = D1_tree-D_predict
The prediction error data D_prescription_error (= Err _{i + k} ) is acquired by executing the process (subtraction process) corresponding to. Then, the subtractor 24 outputs the acquired prediction error data D_prescription_error (= Erri _{+ k} ) to the prediction error distribution data acquisition unit 254 of the prediction error distribution acquisition unit 25.

Prediction error distribution data acquisition unit 254, a data D_prediction_error output from the subtractor 24, data D3 output from Rarity classifying unit _{253 (= j = class (r} i), class (x) is Rarity value x A function to get the number of the class to which it belongs) and. The prediction error distribution data acquisition unit 254 distributes the data D_prescription_error to classes based on the data D3, and acquires the occurrence frequency distribution of the data D_prescription_error for each distributed class (for the data group distributed to the same class). Then, the prediction error distribution data (Err_dist _j ) is acquired. Then, the prediction error distribution data acquisition unit 254 _{stores the data D4 including the acquired prediction error distribution data Err_dist j} in the second storage unit St3.

In general, the prediction error classified into class 0 (Class_0) corresponds to a normal input pattern (time-series data of frequently occurring QoS values), so that the value is small and the prediction is classified into class 0 (Class_0). The dispersion (or standard deviation) of the distribution of the error data group is small.

On the other hand, the prediction error classified into class 1 (Class_1) has a lower frequency of occurrence than the input pattern of class 0, so its value is larger than the prediction error classified into class 0, and the value is larger than the prediction error classified into class 0 (Classs_1). ), The dispersion (or standard deviation) of the distribution of the forecast error data group classified in class 0 (Class_0) is larger than the dispersion (or standard deviation) of the distribution of the forecast error data group classified in class 0 (Class_0).

Further, since the prediction error classified into class 2 (Classs_1) has a lower frequency of appearance than the input pattern of class 1, its value is larger than the prediction error classified into class 1 and is classified into class 2 (Classs_2). The variance (or standard deviation) of the distribution of the forecast error data groups classified is greater than the variance (or standard deviation) of the distribution of the forecast error data groups classified in class 1 (Class_1).

Due to the above characteristics, the prediction error distribution data Err_dist _{0 of the prediction error classified into class 0 (Class_0), the prediction error distribution data Err_dist 1} of the prediction error classified into class 1 (Class_1), and the class. _{The prediction error distribution data Err_dist 2} of the prediction error classified into 2 (Class_2) has, for example, a distribution as shown in the middle of the right end of FIG.

(1.2.2: Prediction processing)
Next, the prediction process (communication failure prediction execution process) will be described.

FIG. 7 is a diagram for explaining a prediction process (a communication failure prediction execution process).

In the prediction process (communication failure prediction execution process), the radio condition prediction device 100 has the prediction value D' _{i + k} of the QoS value at the future time and the prediction error distribution data D4 (=) of the class corresponding to the prediction value D' _{i + k.} Based on Err_dist _j ), a risk value indicating the possibility of a communication failure is acquired at the time (future time) when the true value of the data D_predicted is acquired.

The demodulation processing unit Dm1 of the radio condition prediction device 100 executes demodulation processing on the radio signal received via the antenna Ant1 and acquires an RF demodulation signal and a BB demodulation signal. The observation data acquisition unit FE1 acquires a QoS value from the RF demodulation signal and / or the BB demodulation signal acquired by the demodulation processing unit Dm1. Then, the observation data acquisition unit FE1 outputs the data including the acquired QoS value as the data Din to the data holding unit 21 of the prediction determination processing unit 2.

The control unit CPU 1 sets the mode signal Mode to a signal value indicating the “prediction processing execution mode” and outputs the mode signal to the prediction determination processing unit 2. Further, the control unit CPU 1 sets the control signal ctl1 to a signal value for executing the "prediction processing" (a signal value at which the terminal 2 of the selector SEL1 is selected), and outputs the control signal ctl1 to the prediction determination processing unit 2.

The data holding unit 21 stores and holds the data Din from the observation data acquisition unit FE1 for a predetermined period. The data holding unit 21 outputs the time-series data of the data Din as the data D1s to the rarity acquisition unit 22 and the predicted value acquisition unit 23. When data Din at time i and is denoted as _{D i,} the data D1s, for example, time-series data _{D1s = {D i, D i} + 1, D i + 2} three consecutive set as. For convenience of explanation, as shown in FIG. 8, the observed value series (time series data) of the QoS values after the time t3 are q ₀ , q ₁ , q ₂ , .... In this case, the data holding unit 21 outputs the data D1s = {q ₀ , q ₁ , q ₂ } to the rarity acquisition unit 22 and the predicted value acquisition unit 23.

The norm acquisition unit 221 of the rarity acquisition unit 22 reads the learned data D_trained Rarity calculation data set Dset2 from the learned data storage unit St1 and inputs the data D1s output from the data holding unit 21. Here, a case where the norm acquisition unit 221 processes the data D1s = {q ₀ , q ₁ , q ₂ } will be described.

The norm acquisition unit 221 calculates the norm (for example, Euclidean distance) by using each row vector of the data set Dset2 for Rarity calculation and the data D1s (= {q ₀ , q ₁ , q _2}). In other words, the norm acquiring unit 221, when the m-th row of the row vectors of Rarity calculation data set Dset2 and Dset2_r _m,
d ₀ = Norm (Dset2_r ₀ , D1s)
d ₁ = Norm (Dset2_r ₁ , D1s)
d ₂ = Norm (Dset2_r ₂ , D1s)
・ ・ ・
d _n1-1 = Norm (Dset2_r _n1-1 , D1s)
Norm (v1, v2): A function to acquire the norms (for example, Euclidean distance) of the vectors v1 and v2.
The set data of the norm {d ₀ , d ₁ , ..., d _n1-1 } is acquired by the process corresponding to. Then, the norm acquisition unit 221 outputs the data including the acquired set data of the norms {d ₀ , d ₁ , ..., d _n1-1 } as the data D_norm to the Rarity calculation unit 222.

The Rarity calculation unit 222 calculates the rarity indicating the rarity of the appearance of the time series data D1s from the data D_norm output from the norm acquisition unit 221. Specifically, the Rarity calculation unit 222
_{r i = min (d i,} d i + 1, ···, d i + n1-1)
min (): By performing the process corresponding to the function to obtain the minimum value of the element, and acquires the time series data _{{D i, D i + 1} , D i + 2} rarity value _{r i} corresponding to the QoS value. Here, since the time series data of the QoS value {D _i , _{Di + 1} , _{Di + 2} } = {c ₀ , c ₁ , c ₂ }, the Rarity calculation unit 222 is used.
r ₀ = min (d ₀ , d ₁ , ..., d _n1-1 )
To get.

Then, Rarity calculating unit 222, the data including the calculated rarity _{r i} as data D2, and outputs to the selector SEL1 of the prediction error distribution obtaining unit 25.

The selector SEL1 selects the terminal 2 according to the control signal ctl1 set in the signal value for executing the "prediction processing" (the signal value at which the terminal 2 of the selector SEL1 is selected), and the input data D2 ( = outputs the _{r i)} in the class determination unit 255.

Class determination unit 255 inputs the data D2 and (= _{r i)} output from the selector SEL1, the aggregate data Th threshold output from the first storage unit St2. Class determination unit 255, the class data D2 to (= _{r i)} and classification based on the aggregate data Th threshold, the data D2 (= _{r i)} is sorted (e.g., the class number j = class _(r i) ). The class determination section 255 outputs the identified class number _{j (= class (r i)} ) to the risk value acquisition unit 26.

The predicted value acquisition unit 23 reads the trained data D_trained (Rarity calculation data set Dset2 and label (correct answer data) data set Dset_Label) from the trained data storage unit St1. The predicted value acquisition unit 23 has the same time-series data D1s (= {q ₀ , q ₁ , q ₂ }) of the QoS value and each row vector of the data set Dset2 for Rarity calculation, as in the process executed by the norm acquisition unit 221. The norm (for example, Euclidean distance) is calculated, the row vector having the minimum value is detected, and the label (correct answer data) corresponding to the detected row vector is specified by using the data set Dset_Label. As a result, the predicted value acquisition unit 23 acquires the label corresponding to the row vector that takes the minimum value of the norm as the predicted data value D_predict of the future time. For example, in the case shown in FIG. 7, the time-series data D1s (= {q ₀ , q ₁ , q _{2 }) of the QoS value and the vector ({a 2} , a ₃ ) in the third row of the data set Dset2 for Rarity calculation. , A ₄ })), the predicted value acquisition unit 23 corresponds to the vector ({a ₂ , a ₃ , a ₄ }) in the third row of the data set Dset2 for Rarity calculation. acquired as the predicted data value D_predict value _{a 6} of the label (label (data set first in the third row data of Dset_Label correct answer data)) of future time (future time i + 4 When the current time i) to.

_{Then, the predicted value acquisition unit 23 outputs the predicted data value D_predict (= D'i + k} = a ₆ ) of the future time (future time i + 4 when the current time i is assumed) acquired as described above to the risk value acquisition unit 26. do.

The risk value acquisition unit 26 is output from the data D_predicted (= D' _{i + k} = a ₆ ) output from the predicted value acquisition unit 23, the class number j output from the class determination unit 255, and the second storage unit St3. Data D4 (= Err_dist _j ), data QoS_req indicating a required reference value of communication quality data, and a mode signal Mode output from the control unit CPU 1 are input. When the mode signal Mode is a signal indicating the "prediction processing execution mode", the risk value acquisition unit 26 of the data D_predicted (= D' _{i + k} = a _{6 ) based on the prediction error distribution data D4 (= Err_dist j} ). At the time (future time) when the true value (q ₄ ) is acquired, the risk value indicating the possibility of communication failure is acquired. Specifically, the risk value acquisition unit 26 reads _{the prediction error distribution data Err_dist j} of the class j determined when acquiring the _{predicted value D_predicted (= D'i + k} ) from the second storage unit St3, and reads the predicted value. The prediction error distribution centered on D_predicted (= D' _{i + k) is acquired.} Then, in the prediction error distribution centered on the predicted value D_predicted (= D' _{i + k} ), the predicted error value in the region below the required reference value QoS_req of the communication quality data is integrated (or integrated), and the integrated value (integrated value) is used. As a risk value. For example, the area of the gray area (the area below the required reference value QoS_req of the communication quality data) in the figure shown near the center of FIG. 7 is the risk value.

Then, the risk value acquisition unit 26 outputs the risk value acquired as described above as data Dout. By considering the risk value acquired by the risk value acquisition unit 26, it is possible to grasp the possibility of a communication failure occurring at a future time, and as a result, it is possible to realize highly accurate prediction processing of the occurrence of a communication failure. ..

As described above, the radio condition prediction device 100 executes the prediction process.

Here, using FIG. 8, when an input pattern (input pattern having a low Rarity value) in which the generation pattern of time-series data of observation data (QoS value) frequently appears is input to the radio condition prediction device 100 (FIG. 9 (FIG. 9). a)) and an input pattern (input pattern having a high Rarity value) in which the generation pattern of time-series data of observation data (QoS value) is rare is input to the radio condition prediction device 100 (FIG. 9 (b)). The acquisition of risk values will be explained.

In the case of FIG. 9A, it is an input pattern (input pattern having a low Rarity value) in which the occurrence pattern of time-series data of the observation data (QoS value) frequently appears, because the Rarity value is low and the frequency of occurrence of the input pattern is high. , The variation (standard deviation, dispersion) _{of the prediction error distribution data Err_dist j} of the class determined by the Rarity value calculated from the input pattern is small. In this case, for example, the class (Class_0) with the class number j = 0 is selected, and the prediction error distribution data Err_dist ₀ is selected. Then, the risk value acquisition unit 26 acquires and acquires a prediction error distribution in which _{the center of the prediction error distribution data Err_dist 0} is the predicted value D' _{i + k} (D' _{i + k} : predicted value of the QoS value at time ti _{+ k).} In the prediction error distribution, the integrated value (integrated value) of the region where the QoS value is equal to or less than the value QoS_req (region AR1 in the case of FIG. 9A) is acquired as a risk value. In the case of FIG. 9A, since the area of the region AR1 is considerably small, the risk value is also small. That is, in the case of FIG. 9A, it is appropriately determined that the possibility of a communication failure occurring at _{time ti + k is considerably small.}

On the other hand, in the case of FIG. 9B, the generation pattern of the time-series data of the observation data (QoS value) is a rare input pattern (input pattern having a high Rarity value), the Rarity value is high, and the frequency of occurrence of the input pattern is high. Since it is low, the variation (standard deviation, dispersion) _{of the prediction error distribution data Err_dist j} of the class determined by the Rarity value calculated from the input pattern is large. In this case, for example, the class (Class_2) with the class number j = 2 is selected, and the prediction error distribution data Err_dist ₂ is selected. Then, the risk value acquisition unit 26 acquires and acquires a prediction error distribution in which _{the center of the prediction error distribution data Err_dist 2} is the predicted value D' _{i + k} (D' _{i + k} : the predicted value of the QoS value at time ti _{+ k).} In the prediction error distribution, the integrated value (integrated value) of the region where the QoS value is equal to or less than the value QoS_req (region AR2 in the case of FIG. 9B) is acquired as a risk value. In the case of FIG. 9B, since the area of the region AR1 is large, the risk value is also large. That is, in the case of FIG. 9A, it is appropriately determined that there is a high possibility that a communication failure will occur at _{time ti + k.}

By processing in this way, the wireless condition prediction device 100 appropriately causes a communication failure even when an input pattern in which the time-series data generation pattern of the observation data (QoS value) is rare is input. It is possible to grasp the possibility of presence or absence and perform highly accurate prediction processing of the occurrence of communication failure.

As described above, in the radio condition prediction device 100, by learning the generation pattern of the time-series data of the observation data (QoS value) at the time of learning, the generation pattern of the time-series data of the observation data (QoS value) that frequently appears. To grasp. Further, in the radio condition prediction device 100, the similarity between the generation pattern of the time-series data of the observation data (QoS value) acquired at the time of learning and the time-series data of the actual observation data (QoS value) is, for example, norm ( For example, it is determined by calculating (Euclidean distance), and a rarity value (Rarity value) indicating the degree of rarity of appearance of time-series data of observation data (QoS value) is calculated. Then, the radio condition prediction device 100 acquires the rarity distribution, which is the distribution of the rarity values, and classifies the data group of the rarity values into a predetermined class according to the threshold value based on the rarity distribution.

Then, the radio situation prediction apparatus 100, at the time of the prediction, the prediction error based on the rarity value r _i of the time-series data of observation data to be processed (QoS values), to identify the class to be classified, corresponding to the class Based on the distribution data Err_dist _j and the predicted value of the future time, the risk value indicating the occurrence probability of the communication failure at the future time is acquired. Then, the radio condition prediction device 100 can appropriately grasp the possibility that a communication failure will occur at a future time based on the acquired risk value. In the wireless condition prediction device 100, the risk value can be acquired as continuous value or multi-value data by processing as described above, so that the determination is based on binary data indicating the presence or absence of a communication failure as in the prior art. Compared to processing, it is possible to realize much more accurate prediction processing.

As described above, in the wireless condition prediction device 100, communication is performed while considering the rarity of the input pattern (time-series data of observation data (QoS value)), which is an important factor when detecting a communication failure. Since it is possible to acquire a risk value indicating the possibility of a failure, it is possible to realize highly accurate prediction processing of the occurrence of a communication failure.

Further, since the wireless condition prediction device 100 executes a prediction process for the occurrence of a communication failure in consideration of the rarity of the input pattern (time-series data of the observation data (QoS value)), a rare input pattern, Alternatively, even when an unlearned input pattern occurs, a risk value indicating the possibility of a communication failure can be appropriately acquired, and as a result, highly accurate prediction processing of the communication failure can be realized. ..

[Second Embodiment]
Next, the second embodiment will be described. The same parts as those in the above embodiment are designated by the same reference numerals, and detailed description thereof will be omitted.

FIG. 9 is a schematic configuration diagram of the radio condition prediction device 200 of the second embodiment.

As shown in FIG. 9, the radio condition prediction device 200 of the second embodiment has a learning unit 1, a learned data storage unit St1, a prediction determination processing unit 2, and a control of the radio condition prediction device 100 of the first embodiment. It has a configuration in which two sets of CPU1 units are provided. That is, as shown in FIG. 9, the radio condition prediction device 200 of the second embodiment has an antenna Ant1, a demodulation processing unit Dm1, an observation data acquisition unit FE1, a first learning prediction unit Dev1, and a second learning prediction. It is provided with a unit Dev2.

The first learning prediction unit Dev1 includes a first learning unit 1A, a first learned data storage unit St1A, a first prediction determination processing unit 2A, and a first control unit CPU1A. The first learning unit 1A, the first learned data storage unit St1A, the first prediction determination processing unit 2A, and the first control unit CPU 1A are the learning unit 1 and the learning unit 1 of the wireless situation prediction device 100 of the first embodiment, respectively. It is the same functional unit as the completed data storage unit St1, the prediction determination processing unit 2, and the control unit CPU1.

The second learning prediction unit Dev2 includes a second learning unit 1B, a second learned data storage unit St1B, a second prediction determination processing unit 2B, and a second control unit CPU1B. The second learning unit 1B, the second learned data storage unit St1B, the second prediction determination processing unit 2B, and the second control unit CPU1B are the learning unit 1 and the learning unit 1 of the wireless situation prediction device 100 of the first embodiment, respectively. It is the same functional unit as the completed data storage unit St1, the prediction determination processing unit 2, and the control unit CPU1.

FIG. 10 is a timing chart showing time-series data (example) of QoS values acquired by the radio condition prediction device 200, a processing mode of the first learning prediction unit Dev1, and a processing mode of the second learning prediction unit Dev2. be. In the timing chart of FIG. 10, the horizontal axis is time.

In the wireless situation prediction device 200, as shown in FIG. 10, the learning process and the prediction process are performed by setting the processing mode of the first learning prediction unit Dev1 and the processing mode of the second learning prediction unit Dev2 to different processing modes. And are executed by pipeline processing. That is, in the wireless situation prediction device 200, the first learning prediction unit Dev1 and the second learning prediction unit Dev2 execute the following processes, and further execute the following processes repeatedly.
(1) During the period from time t00 to t10, the first learning prediction unit Dev1 executes the learning process (the same process as the learning process of the first embodiment). Then, the first control unit CPU1A outputs a mode signal ModeA having a signal value indicating the "learning mode" during this period.
(1-1) During the period from time t00 to t01, the first learning prediction unit Dev1 executes the process A11 of the first step (the same process as the process A1 of the first step of the first embodiment). Then, in this period, the first control unit CPU1A sets the control signal ctl1A (the signal corresponding to the control signal ctl1 of the first embodiment) to the signal value for executing the "processing of the first step", and the first 1 Output to the prediction determination processing unit 2A.
(1-2) During the period from time t01 to t02, the first learning prediction unit Dev1 executes the second step process A12 (the same process as the second step process A2 of the first embodiment). Then, the first control unit CPU1A sets the control signal ctl1A to a signal value for executing the "processing of the second step" in this period, and outputs the control signal ctl1A to the first prediction determination processing unit 2A.
(1-3) During the period from time t02 to t10, the first learning prediction unit Dev1 executes the process A13 of the third step (the same process as the process A3 of the third step of the first embodiment). Then, the first control unit CPU1A sets the control signal ctl1A to a signal value for executing the "processing of the third step" in this period, and outputs the control signal ctl1A to the first prediction determination processing unit 2A.
(2) During the period of time t10 to t20, the first learning prediction unit Dev1 executes the prediction process (the same process as the prediction process of the first embodiment) and outputs the determination result DoutA. Then, the first control unit CPU1A outputs a mode signal ModeA having a signal value indicating the "prediction processing execution mode" during this period.
(2-1) During the period from time t10 to t11, the second learning prediction unit Dev2 executes the process B11 of the first step (the same process as the process A1 of the first step of the first embodiment). Then, the second control unit CPU1B sets the control signal ctl1B (the signal corresponding to the control signal ctl1 of the first embodiment) to the signal value for executing the "processing of the first step" in this period, and the second control unit CPU1B 2 Output to the prediction determination processing unit 2B.
(2-2) During the period from time t11 to t12, the second learning prediction unit Dev2 executes the second step process B12 (the same process as the second step process A2 of the first embodiment). Then, the second control unit CPU1B sets the control signal ctl1B to a signal value for executing the "processing of the second step" in this period, and outputs the control signal ctl1B to the second prediction determination processing unit 2B.
(2-3) During the period from time t12 to t20, the second learning prediction unit Dev2 executes the process B13 of the third step (the same process as the process A3 of the third step of the first embodiment). Then, the second control unit CPU1B sets the control signal ctl1B to a signal value for executing the "processing of the third step" in this period, and outputs the control signal ctl1B to the second prediction determination processing unit 2B.
(3) During the period from time t20 to t30, the second learning prediction unit Dev2 executes the prediction process (the same process as the prediction process of the first embodiment) and outputs the determination result DoutB. Then, the second control unit CPU1B outputs a mode signal ModeB having a signal value indicating the “prediction processing execution mode” during this period.
(3-1) During the period from time t20 to t21, the first learning prediction unit Dev1 executes the process A21 of the first step (the same process as the process A1 of the first step of the first embodiment). Then, in this period, the first control unit CPU1A sets the control signal ctl1A (the signal corresponding to the control signal ctl1 of the first embodiment) to the signal value for executing the "processing of the first step", and the first 1 Output to the prediction determination processing unit 2A.
(3-2) During the period from time t21 to t22, the first learning prediction unit Dev1 executes the second step process A22 (the same process as the second step process A2 of the first embodiment). Then, the first control unit CPU1A sets the control signal ctl1A to a signal value for executing the "processing of the second step" in this period, and outputs the control signal ctl1A to the first prediction determination processing unit 2A.
(3-3) During the period from time t22 to t30, the first learning prediction unit Dev1 executes the process A23 of the third step (the same process as the process A3 of the third step of the first embodiment). Then, the first control unit CPU1A sets the control signal ctl1A to a signal value for executing the "processing of the third step" in this period, and outputs the control signal ctl1A to the first prediction determination processing unit 2A.

In the wireless situation prediction device 200, the learning process and the prediction process can be executed by the pipeline process by performing the process as described above. As a result, since the wireless condition prediction device 200 can always execute the learning process and the prediction process, it is possible to predict the communication failure with higher accuracy.

[Other embodiments]
In the above embodiment, the case where the observed value is a QoS value has been described, but the observed value is not limited to this, and the observed value is, for example, other data as long as it can grasp the situation of the wireless environment. Data may be used.

Further, the learned data storage unit St1, the first storage unit St2, the second storage unit St3, the first learned data storage unit St1A, and the second learned data storage unit St1B are the wireless status prediction devices 100 and 200. It may be provided externally.

Further, in the above embodiment, the case where the prediction error distribution data is acquired by acquiring the occurrence frequency distribution of the data D_prescription_error and the risk value is acquired by using the acquired prediction error distribution data has been described. It is not limited, and may be processed as follows, for example. In the radio condition prediction devices 100 and 200, the average value of the data D_prescription_error is μ _j (μ _j : the average value of the prediction error values classified into the class of the class number j) from the occurrence frequency distribution of the data D_prescription_error for each classified class. ), Standard deviation σ _j (σ _j : standard deviation of the prediction error values classified in the class of class number j), and the prediction error distribution data Err_dist _j follows the normal distribution N (μ _j , σ _j ^2). It may be regarded as a distribution (approximate) and a process of acquiring a risk value may be executed.

Further, in the above embodiment, the case of predicting a communication failure in a wireless environment has been described, but the present invention is not limited to this, and in the wireless status prediction device, for example, a wired communication environment, or wireless or wired can be used. QoS information may be acquired from a hub in the backbone of a mixed network to predict a communication failure in the communication environment.

Further, in the radio condition prediction devices 100 and 200 described in the above embodiment, each block may be individually integrated into one chip by a semiconductor device such as an LSI, or may be integrated into one chip so as to include a part or all of the blocks. May be.

Although it is referred to as LSI here, it may be referred to as IC, system LSI, super LSI, or ultra LSI depending on the degree of integration.

Further, the method of making an integrated circuit is not limited to the LSI, and may be realized by a dedicated circuit or a general-purpose processor. An FPGA (Field Programmable Gate Array) that can be programmed after the LSI is manufactured, or a reconfigurable processor that can reconfigure the connection and settings of the circuit cells inside the LSI may be used.

Further, a part or all of the processing of each functional block of each of the above embodiments may be realized by a program. Then, a part or all of the processing of each functional block of each of the above embodiments is performed by the central processing unit (CPU) in the computer. Further, the program for performing each process is stored in a storage device such as a hard disk or a ROM, and is read and executed in the ROM or in the RAM.

Further, each process of the above embodiment may be realized by hardware, or may be realized by software (including the case where it is realized together with an OS (operating system), middleware, or a predetermined library). Further, it may be realized by mixed processing of software and hardware.

For example, when each functional unit of the above embodiment is realized by software, the hardware configuration shown in FIG. 11 (for example, a hardware configuration in which a CPU, ROM, RAM, input unit, output unit, etc. are connected by a bus Bus). May be used to realize each functional unit by software processing.

Further, the execution order of the processing methods in the above-described embodiment is not necessarily limited to the description of the above-mentioned embodiment, and the execution order can be replaced without departing from the gist of the invention.

A computer program that causes a computer to perform the above-mentioned method and a computer-readable recording medium that records the program are included in the scope of the present invention. Here, examples of computer-readable recording media include flexible disks, hard disks, CD-ROMs, MOs, DVDs, DVD-ROMs, DVD-RAMs, large-capacity DVDs, next-generation DVDs, and semiconductor memories. ..

The computer program is not limited to the one recorded on the recording medium, and may be transmitted via a telecommunication line, a wireless or wired communication line, a network represented by the Internet, or the like.

The specific configuration of the present invention is not limited to the above-described embodiment, and various changes and modifications can be made without departing from the gist of the invention.

100, 200 Wireless situation prediction device 1 Learning unit 2 Prediction judgment processing unit 22 Rarity acquisition unit 23 Prediction value acquisition unit 24 Subtractor (subtraction unit)
25 Prediction error distribution acquisition unit 26 Risk value acquisition unit St1 Learned data storage unit Dev1 First learning prediction unit Dev2 Second learning prediction unit

Claims (8)

(1) In order to make the communication status prediction process feasible during the learning period, the communication quality data and the time-series pattern of the communication quality data are acquired, and the prediction error distribution is acquired from the communication quality data. , (2) A communication status prediction device that performs communication status prediction processing using the prediction error distribution acquired in the learning period during the prediction processing execution period.
In the learning period and the prediction processing execution period, a prediction value acquisition unit that acquires the prediction value of the communication quality data after a predetermined time from the time series pattern of the communication quality data input at the current time, and the prediction value acquisition unit.
The true value of the predicted value of the communication quality data acquired by the predicted value acquisition unit in the learning period and the value of the communication quality data actually observed after the predetermined time from the current time. The prediction error distribution acquisition unit that acquires the prediction error distribution, which is the distribution of the prediction error values that is the difference between
It is a rarity acquisition unit that acquires a rarity value indicating the degree of rarity of the occurrence frequency of the time-series pattern of the communication quality data, which is data indicating the communication quality, and is the time-series pattern of the communication quality data acquired in the learning period. A data set for calculating rarity including an occurrence pattern is acquired, and among the acquired data sets for calculating rarity, the time-series pattern having the smallest difference from the observed value of the communication quality data is the minimum value time-series pattern. The rarity acquisition unit that acquires a value based on the norm of the extracted minimum value time-series pattern and the time-series pattern of the observed value of the communication quality data as the rarity value.
In the prediction processing execution period, (1) the prediction error distribution acquired by the prediction error distribution acquisition unit during the learning period, and (2) the prediction acquired by the prediction value acquisition unit during the prediction processing execution period. Based on the value and (3) the required reference value of the communication quality data, the probability that the predicted value does not satisfy the required reference value of the communication quality data at the future time corresponding to the predicted target time of the predicted value. estimates obtained using the prediction error distribution, the risk value acquiring unit that the estimated value obtained, the future time the communication failure to obtain a risk value indicating the possibility of occurrence,
Equipped with
The prediction error distribution acquisition unit
In the learning period, the range of possible values of the rarity value is classified into a plurality of categories, a plurality of classes corresponding to the range of possible values of the classified rarity values are determined, and the prediction is made for each determined class. Get the error distribution,
In the prediction processing execution period, the class determined by the range including the rarity value acquired by the rarity acquisition unit for the time series pattern of the observed value of the communication quality data is specified.
The risk value acquisition unit
In the prediction processing execution period, the prediction error distribution corresponding to the predicted value and the class determined based on the rarity value acquired by the rarity acquisition unit for the time series pattern of the observed value of the communication quality data. Based on the above, the risk value indicating the possibility of a communication failure at a future time corresponding to the predicted target time of the predicted value is acquired.
Communication status predictor. A learning unit that executes a learning process using the time-series pattern generation pattern of the communication quality data and the communication quality data of the future time after the time-series pattern is generated.
Rarity calculation data set including the generation pattern of the time series pattern of the communication quality data acquired by the learning unit, and the communication quality data of the future time after a predetermined time after the time series pattern is generated. A storage unit that stores trained data, including a label dataset containing
Further prepare
The rarity acquisition unit calculates the norms of the time-series pattern of the communication quality data acquired at the current time and the plurality of time-series patterns of the communication quality data included in the rarity calculation data set. , Acquire the minimum value of the calculated norms as the rarity value,
The communication status prediction device according to claim 1. The prediction error distribution acquisition unit
By acquiring the frequency distribution of the rarity value at a predetermined time, the rarity distribution data is acquired, and the rarity distribution data is acquired.
Acquire a threshold group for classifying the acquired rarity values included in the rarity distribution into a predetermined number of classes.
The communication status prediction device according to claim 2. Further, a subtraction unit for acquiring a prediction error value which is a difference between the predicted value of the communication quality data acquired by the predicted value acquisition unit and the true value of the predicted value is provided.
The prediction error distribution acquisition unit
In the learning period, the prediction error value is classified into a class determined based on the rarity value and the threshold group of the time series pattern of the communication quality data used when acquiring the prediction error value. , Acquire the prediction error distribution data of the prediction error value for each classified class,
The risk value acquisition unit
In the prediction processing execution period, when the center of the prediction error distribution data corresponding to the class determined by the rarity value of the time series pattern is set to the prediction value, it is equal to or less than the required reference value of the communication quality data. The value obtained by integrating the regions is acquired as the risk value.
The communication status prediction device according to claim 3. The prediction error distribution acquisition unit
Set the prediction error distribution data for each classified class to a normal distribution determined by the mean and variance of the prediction error values belonging to the class.
The communication status prediction device according to claim 4. The first learning prediction unit and
The second learning prediction unit and
It is a communication status prediction device equipped with
The first learning prediction unit is
(1) In the first learning period, in order to make the communication status prediction process executable, the communication quality data and the time-series pattern of the communication quality data are acquired, and the prediction error distribution is obtained from the communication quality data. It is a functional unit that performs prediction processing of communication status by using the prediction error distribution acquired in (2) the first prediction processing execution period and the prediction error distribution acquired in the first learning period.
A first learning unit that executes learning processing using the time-series pattern generation pattern of communication quality data and the communication quality data at a future time after the time-series pattern is generated.
Rarity calculation data set including the generation pattern of the time series pattern of the communication quality data acquired by the first learning unit, and the communication at a future time after a predetermined time after the time series pattern is generated. A first trained data storage unit that stores trained data including a label data set containing quality data, and a first trained data storage unit.
It is the first prediction judgment processing unit that performs the prediction judgment processing of the occurrence of communication failure.
In the first learning period and the first prediction processing execution period, the first prediction that acquires the predicted value of the communication quality data after a predetermined time from the time series pattern of the communication quality data input at the current time. Value acquisition part and
In the first learning period, the predicted value of the communication quality data acquired by the predicted value acquisition unit and the predicted value which is the value of the communication quality data actually observed after the predetermined time from the current time. The first prediction error distribution acquisition unit that acquires the prediction error distribution, which is the distribution of the prediction error values that is the difference from the true value,
In the first prediction processing execution period, (1) the prediction error distribution acquired by the prediction error distribution acquisition unit during the first learning period, and (2) acquisition by the prediction value acquisition unit during the prediction processing execution period. Based on the predicted value and (3) the required reference value of the communication quality data, the predicted value sets the required reference value of the communication quality data at a future time corresponding to the predicted target time of the predicted value. an estimate of the probability that do not satisfy obtained using the prediction error distribution, and acquired the estimated value, the first risk value acquiring unit that future time the communication failure to obtain a risk value indicating the possibility of occurrence,
The first prediction determination processing unit including
Equipped with
The second learning prediction unit is
(1) In the second learning period, in order to make the communication status prediction process executable, the communication quality data and the time-series pattern of the communication quality data are acquired, and the prediction error distribution is obtained from the communication quality data. It is a functional unit that performs prediction processing of communication status by using the prediction error distribution acquired in (2) the second learning period in the second learning period.
A second learning unit that executes a learning process using the time-series pattern generation pattern of the communication quality data and the communication quality data of the future time after the time-series pattern is generated.
The data set for rarity calculation including the generation pattern of the time series pattern of the communication quality data acquired by the second learning unit, and the communication at a future time after a predetermined time after the time series pattern is generated. A second trained data storage unit that stores trained data including a label data set containing quality data, and a second trained data storage unit.
It is the second prediction judgment processing unit that performs the prediction judgment processing of the occurrence of communication failure.
In the second learning period and the second prediction processing execution period, the second prediction that acquires the predicted value of the communication quality data after a predetermined time from the time series pattern of the communication quality data input at the current time. Value acquisition part and
In the second learning period, the predicted value of the communication quality data acquired by the predicted value acquisition unit and the predicted value which is the value of the communication quality data actually observed after the predetermined time from the current time. The second prediction error distribution acquisition unit that acquires the prediction error distribution, which is the distribution of the prediction error values that is the difference from the true value,
In the first prediction processing execution period, (1) the prediction error distribution acquired by the prediction error distribution acquisition unit during the first learning period, and (2) acquisition by the prediction value acquisition unit during the prediction processing execution period. Based on the predicted value and (3) the required reference value of the communication quality data, the predicted value sets the required reference value of the communication quality data at a future time corresponding to the predicted target time of the predicted value. an estimate of the probability that do not satisfy obtained using the prediction error distribution, and acquired the estimated value, the second risk value acquiring unit that future time the communication failure to obtain a risk value indicating the possibility of occurrence,
The second prediction determination processing unit including
Equipped with
The first learning prediction unit is
(1) When the second learning prediction unit is executing the prediction determination process, the learning process is executed and the learning process is executed.
(2) When the second learning prediction unit is executing the learning process, the prediction determination process is executed.
Communication status predictor. (1) In order to make the communication status prediction process feasible during the learning period, the communication quality data and the time-series pattern of the communication quality data are acquired, and the prediction error distribution is acquired from the communication quality data. , (2) A communication status prediction method for predicting a communication status using the prediction error distribution acquired in the learning period during the prediction processing execution period.
In the learning period and the prediction processing execution period, a prediction value acquisition step of acquiring the prediction value of the communication quality data after a predetermined time from the time series pattern of the communication quality data input at the current time, and the prediction value acquisition step.
In the learning period, the predicted value of the communication quality data acquired by the predicted value acquisition step and the true value of the predicted value which is the value of the communication quality data actually observed after the predetermined time from the current time. The prediction error distribution acquisition step to acquire the prediction error distribution, which is the distribution of the prediction error values, which is the difference between
It is a rarity acquisition step for acquiring a rarity value indicating the degree of rarity of the occurrence frequency of the time-series pattern of the communication quality data, which is data indicating the communication quality, and is a step of acquiring the time-series pattern of the communication quality data acquired in the learning period. A data set for calculating rarity including an occurrence pattern is acquired, and among the acquired data sets for calculating rarity, the time-series pattern having the smallest difference from the observed value of the communication quality data is the minimum value time-series pattern. The rarity acquisition step of acquiring the value based on the norm of the extracted minimum value time-series pattern and the time-series pattern of the observed value of the communication quality data as the rarity value.
In the prediction processing execution period, (1) the prediction error distribution acquired by the prediction error distribution acquisition step during the learning period, and (2) the prediction acquired by the prediction value acquisition step during the prediction processing execution period. Based on the value and (3) the required reference value of the communication quality data, the probability that the predicted value does not satisfy the required reference value of the communication quality data at the future time corresponding to the predicted target time of the predicted value. estimates obtained using the prediction error distribution, the estimated value obtained, the risk value obtaining step in which the future time the communication failure to obtain a risk value indicating the possibility of occurrence,
Equipped with
In the prediction error distribution acquisition step,
In the learning period, the range of possible values of the rarity value is classified into a plurality of categories, a plurality of classes corresponding to the range of possible values of the classified rarity values are determined, and the prediction is made for each determined class. Get the error distribution,
In the prediction processing execution period, the class determined by the range including the rarity value acquired by the rarity acquisition step for the time series pattern of the observed value of the communication quality data is specified.
In the risk value acquisition step,
In the prediction processing execution period, the prediction error distribution corresponding to the predicted value and the class determined based on the rarity value acquired by the rarity acquisition step for the time series pattern of the observed value of the communication quality data. Based on the above, the risk value indicating the possibility of a communication failure at a future time corresponding to the predicted target time of the predicted value is acquired.
Communication status prediction method. A program for causing a computer to execute the communication status prediction method according to claim 7.

Images