JP7699359B2 - Communication quality prediction model update device, communication quality prediction model update method, and communication quality prediction model update program - Google Patents

Description

The present invention relates to a communication quality prediction model update device, a communication quality prediction model update method, and a communication quality prediction model update program.

As traffic on the Internet, IoT (Internet of Things), and M2M (Machine to Machine) increases, the use of higher frequencies is being considered. Next-generation mobile communications, 5G and 6G, are expected to realize high-speed, large-capacity communications using frequencies above 30 GHz, known as millimeter waves.

On the other hand, wireless communications using high frequencies such as sub6 GHz, millimeter waves, and terahertz waves are strongly affected by the surrounding environment, and it is known that communication quality drops sharply when the wireless communication path is blocked by the human body or other objects (Non-Patent Document 1).

Such abrupt changes in communication quality can significantly reduce the perceived quality of communication. Therefore, there is a need for a device that can predict wireless channel blockages in advance and control communication to reduce their impact.

For example, a device is known that uses an RGB camera, a depth camera, and LiDAR (Light Detection and Ranging) to acquire physical space information such as depth images and point cloud information of a wireless communication area, and uses the physical space information as input to learn the correspondence with the communication quality value of wireless communication through machine learning, predicts future communication quality in advance using a communication quality prediction model that maps the physical space information to a communication quality value, and performs handover control and transmission power control based on the prediction results (Non-Patent Document 2).

S. Collonge and others, “Influence of the human activity on wide-band characteristics of the 60 GHz in-door radio channel”, IEEE Transactions. Wireless Communications, Vol. 3, No. 6, November 2004, p. 2396-p. 2406 T. Nishio and 7 others, “Proactive Received Power Prediction Using Machine Learning and Depth Images for mmWave Networks”, IEEE Journal on Selected Areas in Communications, Vol.37, No.11, November 2019, p.2413-p.2427

However, in Non-Patent Document 2, it is assumed that the communication quality prediction model that has been trained by machine learning will be repeatedly used, and therefore there is a problem that as time passes, differences gradually arise between the time of machine learning and the current state of the communication environment, resulting in a decrease in the accuracy of communication quality prediction.

The present invention has been made in consideration of the above circumstances, and the object of the present invention is to provide a technology that can improve the accuracy of predicting future communication quality in a wireless communication area.

A communication quality prediction model update device according to one embodiment of the present invention includes a first acquisition unit that acquires physical space information, which is physical information within a wireless communication area; a second acquisition unit that acquires communication information of currently wireless communication within the wireless communication area; a first storage unit that stores the physical space information and the communication information; a second storage unit that stores a communication quality prediction model for predicting future communication quality related to wireless communication within the wireless communication area; a determination unit that determines an update frequency or update timing of the communication quality prediction model; and an update unit that updates the communication quality prediction model using the physical space information and the communication information based on the update frequency or the update timing.

A communication quality prediction model update method according to one aspect of the present invention is a communication quality prediction model update method performed by a communication quality prediction model update device, which includes the steps of acquiring physical space information, which is physical information within a wireless communication area, acquiring communication information of currently wireless communication within the wireless communication area, storing the physical space information and the communication information, determining an update frequency or update timing of a communication quality prediction model for predicting future communication quality related to wireless communication within the wireless communication area, and updating the communication quality prediction model using the physical space information and the communication information based on the update frequency or the update timing.

The communication quality prediction model update program of one aspect of the present invention causes a computer to function as the communication quality prediction model update device.

The present invention provides technology that can improve the accuracy of predicting future communication quality in a wireless communication area.

FIG. 1 is a diagram showing the overall configuration of the system. FIG. 2 is a flowchart showing the first updating method. FIG. 3 is a flowchart showing the second updating method. FIG. 4 is a diagram showing a prediction error of communication quality. FIG. 5 is a diagram illustrating a hardware configuration of the communication quality prediction model update device.

Below, an embodiment of the present invention will be described with reference to the drawings. In the description of the drawings, the same parts are given the same reference numerals and the description will be omitted.

Summary of the Invention
The present invention discloses a technique for updating a communication quality prediction model. Specifically, the technique obtains physical space information included in a wireless communication area and communication information of currently wireless communication, and updates the communication quality prediction model based on the obtained information.

At this time, the update frequency or update timing is set (updated) based on the relative positions of the camera or LiDAR installation location and the wireless communication access point installation location, reducing the calculation costs and data collection costs required to update the communication quality prediction model.

In the present invention, the communication quality prediction model is updated, which makes it possible to avoid a situation in which a discrepancy occurs between the predicted value of communication quality based on the communication quality prediction model and the current actual measured value over time, thereby improving prediction accuracy.

[Overall system configuration]
1 is a diagram showing the overall configuration of a system according to this embodiment. The system according to this embodiment includes a communication quality prediction model update device 1, a communication quality prediction device 2, and a communication control device 3.

The communication quality prediction device 2 is a device that predicts the future communication quality of each wireless communication path R formed between an access point 4 that performs wireless communication in a wireless communication area AR and each user terminal 5, using a communication quality prediction model and physical space information obtained from a camera 6. Here, the camera 6 may be a sensor such as LiDAR.

The communication control device 3 is a device that controls communication of each wireless communication channel R based on the prediction results of the future communication quality of each wireless communication channel R predicted by the communication quality prediction device 2. For example, the communication control device 3 performs handover control and transmission power control for a wireless communication channel R whose communication quality is predicted to be degraded due to obstruction by a human body or the like.

The communication quality prediction model update device 1 is a device that updates the communication quality prediction model used by the communication quality prediction device 2. Specifically, the communication quality prediction model update device 1 acquires physical space information contained in the wireless communication area AR and communication information of currently wireless communication, appropriately updates the communication quality prediction model based on that information, and transmits the updated communication quality prediction model to the communication quality prediction device 2.

For example, as shown in FIG. 1, the communication quality prediction model update device 1 includes a physical space information acquisition unit 11, a communication information acquisition unit 12, a data storage unit 13, a communication quality prediction model storage unit 14, an update rule determination unit 15, and a model update unit 16.

The physical space information acquisition unit (first acquisition unit) 11 has a function of acquiring physical space information related to physical information such as the position and existence of objects within the wireless communication area AR as image data or point cloud data from a camera 6 or a sensor installed near the wireless communication area AR.

For example, the physical space information acquisition unit 11 acquires image data of physical space information captured by an RGB camera or a depth camera, and point cloud data of physical space information detected by LiDAR, etc., from the camera 6 in a time series. The RGB camera, the depth camera, and the LiDAR are examples of collection means for collecting physical space information within the wireless communication area AR.

Physical space information includes, for example, dynamic objects such as moving people, cars, and communication devices, growing plants, weather, buildings under construction, and rivers with fluctuating water levels, as well as static objects such as luggage, benches, traffic lights, etc.

In addition, since the point cloud data obtained by the LiDAR is information on the reflection points of light as seen by the LiDAR, the physical space information acquisition unit 11 may convert the point cloud data of the physical space information obtained from the LiDAR into physical space information in a coordinate system that can be recognized by the communication quality prediction model update device 1.

For example, the physical space information acquisition unit 11 may perform signal processing such as conversion processing to convert the obtained physical space information into a multidimensional coordinate system or voxels, normalization of values, and extraction of difference information, before outputting the information to the data storage unit 13. A voxel is a data format that represents a solid object, and by assigning values (e.g., 0 or 1, or coefficients between 0 and 1) to cubes that divide three-dimensional space into a grid, it represents whether an object exists in a small area of the cube, and the characteristics of what exists.

The communication information acquisition unit (second acquisition unit) 12 has a function of acquiring communication information of the current wireless communication between an access point 4 installed within the wireless communication area AR and each user terminal 5 from the access point 4. The access point 4 and the user terminal 5 are examples of communication devices (transmitting/receiving station, transmitting station, receiving station).

The communication information may be, for example, received signal power, signal-to-noise power ratio (SNR), signal-to-interference plus noise power ratio (SINR), received signal strength indication (RSSI), received signal reference quality (RSRQ), packet error rate, number of arriving bits, number of arriving bits per unit time, modulation and coding scheme index (MCS), number of retransmissions, delay time, error correction method, frequency of the communication system, frequency conditions such as bandwidth of the resources used, traffic of the communication network, etc. It may also be differential information of these values, an index calculated by substituting these values into a specified formula, a system setting item that affects the index, etc.

The data storage unit (first storage unit) 13 has a function of storing the physical space information acquired by the physical space information acquisition unit 11 and the communication information of the currently wirelessly communicated information acquired by the communication information acquisition unit 12. The data storage unit 13 may perform preprocessing such as normalizing the absolute values of the information, or may select data to be saved.

The communication quality prediction model memory unit (second memory unit) 14 has a function of storing a communication quality prediction model for predicting future communication quality related to wireless communication within the wireless communication area AR.

When there are multiple cameras 6, a communication quality prediction model may be prepared for each camera, or one for each of the multiple cameras. When there are multiple cameras 6, for example, when there are multiple cameras at different positions in one wireless communication area, or when there is one or more cameras in each of the multiple wireless communication areas.

The update rule determination unit (determination unit) 15 has a function of determining the update rule for the communication quality prediction model. The update rule is, for example, the update period, the update frequency, the update timing, the update conditions, and the update method. The update rule determination unit 15 may determine the same update rule for multiple or all communication quality prediction models, or may determine a different update rule for each communication quality prediction model.

The model update unit (update unit) 16 has a function of updating the communication quality prediction model stored in the communication quality prediction model storage unit 14 using the physical space information and communication information stored in the data storage unit 13 based on the update rule determined by the update rule determination unit 15.

For example, the model update unit 16 updates the communication quality prediction model currently being used by the communication quality prediction device 2 by performing fine tuning, transfer learning, and the like using newer physical space information and communication information. The updated communication quality prediction model is input to the communication quality prediction device 2 and used for wireless communication, control of the user terminal 5, and the like.

[Method of updating communication quality prediction model]
(First Update Method)
FIG. 2 is a flowchart showing a first method for updating a communication quality prediction model.

Step S101:
First, the physical space information acquisition unit 11 acquires physical space information within the wireless communication area AR in chronological order from the cameras 6 and sensors installed in the vicinity of the wireless communication area AR, and stores the information in the data storage unit 13 .

The communication information acquisition unit 12 also acquires, in chronological order, communication information on the current wireless communication between the access point 4 and each user terminal 5 from the access point 4 installed within the wireless communication area AR, and stores the information in the data storage unit 13.

Step S102:
Next, the model update unit 16 determines whether or not it satisfies the update frequency or update timing defined in the update rule of the update rule determination unit 15. For example, the model update unit 16 determines that the update frequency is satisfied when the current time exceeds the time obtained by adding the update period to the previous update time.

Step S103:
If the update frequency or update timing defined in the update rule is satisfied, the model update unit 16 reads the most recent physical space information and communication information (=Training data) from the data storage unit 13, reads validation data for tuning machine learning parameters from a predetermined database, and updates the current communication quality prediction model by performing machine learning such as fine tuning and transfer learning using the read information and data in accordance with the update conditions and update methods defined in advance in the update rule. After that, the model update unit 16 returns to step S102.

On the other hand, if the update frequency or update timing specified in the update rule is not met, the model update unit 16 repeats step S102 without updating the communication quality prediction model.

(Second Update Method)
FIG. 3 is a flowchart showing a second method for updating the communication quality prediction model.

The first update method described above is limited to updating the communication quality prediction model, so it is unclear whether the communication quality prediction model and update conditions are appropriate. The second update method improves this point.

Step S201 to Step S202
This is the same as steps S101 to S102.

Step S203:
If the update frequency or update timing defined in the update rule is satisfied, the model update unit 16 updates the current communication quality prediction model in the same manner as in step S103, and stores (maintains) the communication quality prediction model before the update.

Step S204:
Next, the model update unit 16 evaluates the updated communication quality prediction model and the communication quality prediction model before the update by using the test data. More specifically, the model update unit 16 uses the test data to calculate the improvement of the communication quality prediction by the updated communication quality prediction model compared to the communication quality prediction model before the update.

The test data is the physical space information and communication information that have been stored in the data storage unit 13 up to this point, and is used to test (evaluate) whether the communication quality prediction model is operating correctly.

For example, the model update unit 16 inputs the same physical space information and communication information into the updated communication quality prediction model and the communication quality prediction model before the update, and calculates the degree to which the accuracy rate of the updated communication quality prediction model has improved compared to the accuracy rate of the prediction results using the communication quality prediction model before the update.

The model update unit 16 may use test data from a long time ago, such as several months, several years, or several decades ago, to test (evaluate) whether the communication quality prediction model has not changed significantly, or may use test data from a few weeks, days, or hours ago to test (evaluate) the performance for the current communication situation. The model update unit 16 may use both past test data and test data from a time relatively close to the past. The communication quality prediction model before the update and past communication quality prediction models can be stored in the communication quality prediction model storage unit 14.

Step S205:
Next, the model update unit 16 evaluates the effect of whether or not the communication quality prediction model before the update has been improved based on the evaluation result of the updated communication quality prediction model, and notifies the update rule determination unit 15 of the evaluation result of the effect. More specifically, the model update unit 16 determines whether or not the improvement degree of the communication quality prediction model after the update calculated in step S204 is equal to or greater than a predetermined improvement degree.

Step S206:
If the updated communication quality prediction model has not been improved from the communication quality prediction model before the update, the update rule determination unit 15 updates (changes) the update conditions, update frequency, update timing, etc. of the model update based on the effectiveness evaluation of the model update input from the model update unit 16.

For example, if the improvement in the prediction result by the updated communication quality prediction model is less than a predetermined improvement, the update rule determination unit 15 updates the update rule so that the update process of the communication quality prediction model is not performed frequently, for example, by lowering the update frequency defined in the update rule or increasing the update timing. Then, the process returns to step S202.

At this time, in conjunction with the update of the update rule by the update rule determination unit 15, the model update unit 16 may discard (delete) the updated communication quality prediction model that has no improvement effect, and transmit the communication quality prediction model before the update that was maintained to the communication quality prediction device 2.

On the other hand, if the updated communication quality prediction model has been improved from the communication quality prediction model before the update, the update rule determination unit 15 does not update the update rule. In this case, if the improvement in prediction accuracy is significant, the update rule determination unit 15 may update the update rule so that the update process of the communication quality prediction model is performed more frequently, such as by increasing the update frequency or reducing the update timing. Then, the process returns to step S202.

Regarding the update rules and timing of steps S102 and S202, the update rule determination unit 15 may determine them based on, for example, a pre-specified time interval, time, accuracy of the communication quality prediction model, the amount or quality of training data for updating the communication quality prediction model, or both. For example, the update rule determination unit 15 can set a pre-specified interval of days, day of the week, date, and time, when the accuracy of the communication quality prediction model falls below the standard accuracy, when a pre-specified amount or more of samples for training data can be obtained, etc.

The first and second update methods have been explained above.

The first and second update methods can be performed for each communication quality prediction model corresponding to each camera (sensor) installed at a different position for one or more wireless communication areas. As confirmed by the experiment described below, the update frequency or update timing required for the communication quality prediction model varies depending on the position of the installed camera (sensor). Therefore, by using the optimal update rule for the camera (sensor) at each position, the prediction results by the communication quality prediction model can be optimized, and the communication quality prediction model can be managed appropriately and efficiently.

In other words, in the communication quality prediction model update device 1, the update rule determination unit 15 determines the update frequency or update timing for each of the multiple communication quality prediction models corresponding to multiple cameras (sensors) installed at different positions or combinations thereof, and the model update unit 16 updates each of the multiple communication quality prediction models based on the frequency or timing determined for each communication quality prediction model.

[How to determine update frequency or update timing]
Next, a method for determining the update frequency or update timing will be described.

The frequency or timing of updating the communication quality prediction model is determined based on the positional relationship between the positions of the camera 6, LiDAR, and other environmental sensors and the positions of the access point 4 and user terminal 5.

Specifically, the straight line connecting the position of the access point 4 and the position of the user terminal 5 is taken as the reference line (ax+by+c=0; a, b, c are arbitrary initial values), and a perpendicular line is drawn from the position of the environmental sensor ( _x1 , _y1 ) to the reference line, and the distance d between the intersection point and the position of the environmental sensor is calculated based on the ``formula for the distance between a point and a line'' in equation (1).

The larger this distance d, the higher the update frequency and the shorter the update timing.

For example, the period during which the communication quality prediction model was not updated and the decrease in the prediction accuracy of the communication quality are measured, and based on the measurement results, an appropriate update frequency is considered according to the distance d, and a correspondence table of the distance d or the range of distance d and the update frequency is created.

To understand the decline in prediction accuracy, the aforementioned test data is prepared for multiple different timings, prediction results for each communication quality are obtained, and the difference (error) from the correct communication quality, which indicates the correct answer, is calculated.

For example, test data is acquired and stored every month, and a communication quality prediction model updated five months ago is used to determine a decline in prediction accuracy based on the change in prediction error based on the communication quality prediction results using monthly test data.

More specifically, if the distance d is within 30 cm and the prediction accuracy starts to decrease to a certain level (e.g., 10%) after one month, the update frequency for "d≦30 cm" is set to "one month." If the distance d is greater than 30 cm and within 1 m, and the prediction accuracy starts to decrease to a certain level (e.g., 10%) after one week, the update frequency for "30 cm < d ≦ 1 m" is set to "one week."

[How to generate a dataset for updating a model]
Next, a method for generating a model updating data set will be described. The model updating data set is a data set based on physical space information and communication information (=training data) and validation data (=validation data).

At the update frequency or update timing determined by the above determination method, a model updating dataset is generated using information from the data storage unit 13. First, N pieces of communication information are extracted in order of most recent time stamp. Next, physical space information corresponding to each of the extracted N pieces of communication information and acquired at a time closest to the time τ seconds before the acquisition time of each piece of communication information is extracted from the data storage unit 13. Then, the N pieces of communication information are associated as labels with these N pieces of physical space information. The reason for associating with physical space information from τ seconds ago is to apply Temporal Difference Labeling (Non-Patent Document 2), which assigns labels by shifting the time in order to predict future communication quality.

Then, a predetermined preprocessing is performed on the extracted communication information and physical space information. This preprocessing is, for example, normalization.

For communication information, weights are multiplied or offsets are added so that the distribution is roughly close to a mean of 0 and a variance of 1. Weights and offsets may be set so that values are distributed between 0 and 1. It may also be important to keep the weights and offset values fixed so that the distribution of values is roughly between -1 and 1, with values that deviate significantly being capped at values such as 1 or -1.

The physical space information is processed to match the input of the communication quality prediction model currently being used. For example, if it is image information, the resolution is adjusted to the same resolution as the input of the communication quality prediction model, and the values are normalized to the range of 0 to 1. If it is point cloud information, outliers are removed and downsampled, and then it is converted to a 2D map or voxels.

The specific steps for point cloud preprocessing are shown below. Point cloud preprocessing is performed by the physical space information acquisition unit 11, the data storage unit 13, or the model update unit 16.

First, outliers (noise points) that are included in the point cloud data during measurement are removed. Removing the noise points makes it possible to achieve accurate predictions. Methods for removing noise points include, for example, calculating the distance from a certain point to each of a certain number of neighboring points and removing points that are greater than a threshold, and removing points that contain a small number of points within a sphere with the point at its center.

Next, the number of points in the point cloud data is reduced (downsampled). Reducing the number of points reduces the number of input dimensions to the communication quality prediction model, and suppresses the amount of calculations required for machine learning. Downsampling methods include, for example, a method of randomly thinning out indexes, and a method of uniformly thinning out indexes at fixed intervals.

Next, the three-dimensional space of the point cloud data is converted into an arbitrary two-dimensional plane. By converting the point cloud data into a two-dimensional plane, the volume of the point cloud data can be further reduced. One method of converting into a two-dimensional plane is, for example, to determine one two-dimensional plane from among multiple two-dimensional planes that form the three-dimensional space of the point cloud data, and map the point cloud data onto that two-dimensional plane.

Next, the point cloud data is converted to voxel data. By converting the point cloud data to voxel data, the volume of the point cloud data can be further reduced. The method of conversion to voxel data involves setting the value of the voxel space to 0 or 1 depending on the number of points in a voxel that divides the three-dimensional space of the point cloud data into a grid. For example, for each voxel, if there is a certain number of points (minimum one) in the space corresponding to the voxel, the value of that voxel is set to 1, and if there are not a certain number of points in that space, the value of that voxel is set to 0.

Another method of converting to voxel data is to set the number of points in the space corresponding to the voxel as the density to the voxel value. For example, if the boundary coordinates of a voxel (the two diagonal points among the eight vertices of a voxel) are ( _x1 , _y1 , _z1 ) and ( _x2 , _y2 , _z2 ), respectively, if there is a point that satisfies " _x1 ≦x≦ _x2 ", " _y1 ≦y≦ _y2 ", or " _z1 ≦z≦ _z2 ", then 1 or the number of points is substituted as the voxel value, and if there is no such point, then 0 is substituted.

Finally, the preprocessed communication information and physical space information that serve as labels are used as the dataset for updating the model.

[Method of updating communication quality prediction model]
The model update unit 16 updates the communication quality prediction model using the model update dataset generated by the above method. Specifically, the model update unit 16 divides the model update dataset into Training data and Validation data, and repeats a process of updating the communication quality prediction model with the Training data and a process of verifying the performance of the communication quality prediction model with the Validation data until no improvement in the model performance (prediction result) is observed.

[Measurement results of prediction error]
4 is a diagram showing prediction errors based on the results of prediction of communication quality by the updated communication quality prediction model. The prediction errors after a certain time has elapsed are shown for two cameras A and B installed at different positions. The distance d, which is the basis for the update frequency or update timing, is larger for camera A.

The communication quality prediction model used is a mathematical model in which processing units that perform linear transformation of inputs are connected in a network shape. We used 3D Convolutional Neural Networks, which are constructed in three dimensions using the Neural Network that is the basis of deep learning and include a Convolution layer that performs convolution.

When the amount of data on the horizontal axis is 0, it represents the prediction error before updating the model. From Figure 4, it can be seen that updating the model reduces the prediction error. It can also be seen that camera A, which has a larger distance d, has a larger prediction error than camera B. In other words, it can be seen that there is a greater decrease in prediction accuracy over time. This tendency also depends on the communication quality prediction model used. It was shown that by updating this communication quality prediction model with a certain amount of data or more, it is possible to reduce the prediction error to a certain level (4 dB in this case) and improve prediction accuracy.

[Industrial applications]
The present invention is a technology for appropriately updating a communication quality model in a system used for predicting communication quality or for control based on communication quality in wireless LAN, particularly in devices equipped with millimeter wave communication functions, to reduce the computational load, the amount of data prepared in a database, and to manage a communication quality model with stable performance over the long term.This invention is particularly useful in applications that require a highly reliable and stable wireless network, such as improving the reliability of wireless communication using high frequencies such as millimeter waves, where communication quality prediction is important.

[Effects of the embodiment]
According to this embodiment, since the communication quality prediction model is updated, it is possible to provide a technique capable of improving the prediction accuracy of future communication quality in a wireless communication area.

In addition, according to this embodiment, the update frequency or update timing is determined for each communication quality prediction model, and multiple communication quality prediction models are updated based on the update frequency or update timing for each communication quality prediction model. This makes it possible to optimize the prediction results using the communication quality prediction models, and to appropriately and efficiently manage the communication quality prediction models.

In addition, according to this embodiment, the update frequency or update timing is determined depending on the degree of improvement in the communication quality prediction accuracy of the updated communication quality prediction model compared to the communication quality prediction model before the update, so that a technology can be provided that can further improve the prediction accuracy of communication quality.

In addition, according to this embodiment, the update frequency or update timing is determined based on the positional relationship between the positions of the cameras and sensors that collect physical space information and the positions of the access points and user terminals that perform wireless communication within the wireless communication area, thereby providing a technology that can further improve the accuracy of predicting communication quality.

[others]
The present invention is not limited to the above embodiment. Various modifications of the present invention are possible within the scope of the gist of the present invention. The communication quality prediction model update device 1 according to the present embodiment can be realized by a computer and a program, and the program can be recorded on a recording medium or provided via a network.

For example, the communication quality prediction model update device 1 according to this embodiment can be realized using a general-purpose computer system including a CPU 901, a memory 902, a storage 903, a communication device 904, an input device 905, and an output device 906, as shown in FIG. 5. The memory 902 and the storage 903 are storage devices. In this computer system, the CPU 901 executes a predetermined program loaded onto the memory 902, thereby realizing each function of the communication quality prediction model update device 1.

The communication quality prediction model update device 1 may be implemented in one computer. The communication quality prediction model update device 1 may be implemented in multiple computers. The communication quality prediction model update device 1 may be a virtual machine implemented in a computer.

The program for the communication quality prediction model update device 1 can be stored in a computer-readable recording medium such as a HDD, SSD, USB memory, CD, or DVD. The program for the communication quality prediction model update device 1 can also be distributed via a communication network.

Reference Signs List 1: Communication quality prediction model update device 11: Physical space information acquisition unit 12: Communication information acquisition unit 13: Data storage unit 14: Communication quality prediction model storage unit 15: Update rule determination unit 16: Model update unit 2: Communication quality prediction device 3: Communication control device 4: Access point 5: User terminal 6: Camera 901: CPU
902: Memory 903: Storage 904: Communication device 905: Input device 906: Output device

Claims (6)

A first acquisition unit that acquires physical space information, which is physical information within a wireless communication area;
a second acquisition unit that acquires communication information of a currently wirelessly communicating device within the wireless communication area;
a first storage unit that stores the physical space information and the communication information;
a second storage unit configured to store a communication quality prediction model for predicting a future communication quality related to wireless communication within the wireless communication area;
A determination unit that determines an update frequency or an update timing of the communication quality prediction model;
an update unit that updates the communication quality prediction model using the physical space information and the communication information based on the update frequency or the update timing,
The determination unit is
Based on a positional relationship between a position of a collecting means for collecting the physical space information and positions of a plurality of communication devices that perform wireless communication within the wireless communication area, the update frequency or the update timing is determined so that the longer the distance between the position of the collecting means and the intersection point when a perpendicular line is drawn from the position of the collecting means to a straight line connecting the positions of the two communication devices, the higher the update frequency or the shorter the update timing.
Communication quality prediction model update device. The communication quality prediction model includes a plurality of communication quality prediction models corresponding to a plurality of collection means installed at different positions that collect the physical space information,
The determination unit determines the update frequency or the update timing for each communication quality prediction model;
The update unit is
The communication quality prediction model update device according to claim 1 , further comprising: updating each of the plurality of communication quality prediction models based on an update frequency or update timing for each of the communication quality prediction models. The determination unit is
3. The communication quality prediction model update device according to claim 1, wherein the update frequency or the update timing is determined according to an improvement in communication quality prediction accuracy of the updated communication quality prediction model compared to the communication quality prediction model before the update. The determination unit is
4. The communication quality prediction model update device according to claim 1 , further comprising: a communication quality prediction model update unit that determines an update frequency or update timing of the communication quality prediction model based on the amount of training data for training the communication quality prediction model. In a communication quality prediction model updating method performed by a communication quality prediction model updating device,
acquiring physical space information, which is physical information within a wireless communication area;
acquiring communication information of currently wirelessly performed communication in the wireless communication area;
storing said physical space information and said communication information;
determining an update frequency or update timing of a communication quality prediction model for predicting future communication quality related to wireless communication in the wireless communication area;
updating the communication quality prediction model using the physical space information and the communication information based on the update frequency or the update timing ;
In the determining step,
Based on a positional relationship between a position of a collecting means for collecting the physical space information and positions of a plurality of communication devices that perform wireless communication within the wireless communication area, the update frequency or the update timing is determined so that the longer the distance between the position of the collecting means and the intersection point when a perpendicular line is drawn from the position of the collecting means to a straight line connecting the positions of the two communication devices, the higher the update frequency or the shorter the update timing.
A method for updating a communication quality prediction model. A communication quality prediction model updating program that causes a computer to function as the communication quality prediction model updating device according to any one of claims 1 to 4 .

Images