JP7799822B2 - System and method for detecting abnormalities in an air conditioning system - Google Patents

Description

The present disclosure relates to a system and method for detecting abnormalities in an air conditioning system.

Configurations for detecting abnormalities in air conditioning systems are known. For example, Japanese Patent Application Laid-Open No. 2017-221023 (Patent Document 1) discloses a fault symptom detection device that accurately estimates the internal state of a compressor by analyzing the q-axis current, which is less susceptible to electrical noise. This fault symptom detection device can improve the accuracy of compressor abnormality detection.

Japanese Patent Application Laid-Open No. 2017-221023

Patent Document 1 discloses that a compressor abnormality is detected when the intensity of the compressor's operating frequency component exceeds a threshold value as a result of FFT (Fast Fourier Transform) analysis. However, since this threshold value can differ depending on the environment in which the air conditioning system is operating, if a common threshold value is used regardless of the environment in which the air conditioning system is operating, the accuracy of detecting an abnormality in the air conditioning system may decrease.

This disclosure has been made to solve the problems described above, and its purpose is to improve the accuracy of abnormality detection in air conditioning systems.

A system according to one aspect of the present disclosure detects an abnormality in an air conditioning system. The system includes a learning device, an inference device, and a determination device. The learning device has trained an inference model that infers normal values of specific parameters of the air conditioning system from the operating data of the air conditioning system using learning data including operating data of the air conditioning system acquired during operating periods in which the air conditioning system is normal out of consecutive first to Nth operating periods (N is a natural number). The inference device uses the inference model to infer normal values from operating data of the air conditioning system acquired during the (N+1)th operating period. The determination device determines whether or not there is an abnormality in the air conditioning system during the (N+1)th operating period based on a comparison between the normal value and the specific parameter during the (N+1)th operating period.

A method according to another aspect of the present disclosure detects an abnormality in an air conditioning system. The method includes the steps of: constructing a trained inference model that infers normal values of specific parameters of the air conditioning system from the operating data of the air conditioning system using learning data including operating data acquired during operating periods in which the air conditioning system is normal among consecutive first through Nth operating periods; inferring normal values from the operating data of the air conditioning system acquired during the (N+1)th operating period using the inference model; and determining whether the air conditioning system is abnormal during the (N+1)th operating period based on a comparison between the normal values and the specific parameters during the (N+1)th operating period.

The system and method disclosed herein can improve the accuracy of detecting abnormalities in an air conditioning system by using an inference model that has been trained using learning data that includes operating data of the air conditioning system acquired during operating periods in which the air conditioning system is normal among the first to Nth consecutive operating periods.

1 is a block diagram showing an example of the configuration of an anomaly detection system according to a first embodiment and an air conditioning system whose state is monitored by the anomaly detection system; FIG. 2 is a functional block diagram showing the configuration of the air conditioning system of FIG. 1. FIG. 2 is a diagram showing an example of operating data that reflects the state of the air conditioning system of FIG. 1 . FIG. 2 is a block diagram showing the configuration of the anomaly detection system 1 of FIG. 1. 2 is a block diagram showing the configuration of the anomaly detection system of FIG. 1 and the flow of data during an Nth operating period (N≧2). FIG. A figure showing an example of a neural network included in the inference model of Figures 4 and 5. A figure showing a confidence interval based on a state index value inferred by a trained inference model and a time chart of the actual state index value. 6 is a flowchart showing an example of an abnormality detection process performed for each operating period in the abnormality detection system of FIGS. 4 and 5 . FIG. 2 is a block diagram showing a hardware configuration of the anomaly detection system shown in FIG. 1 . FIG. 10 is a block diagram showing the configuration of an anomaly detection system according to a second embodiment and the flow of data during a first operating period. FIG. 10 is a block diagram showing the configuration of an anomaly detection system according to a second embodiment and the flow of data during an Nth operating period (N≧2). 12 is a flowchart showing an example of an abnormality detection process performed for each operation period in the abnormality detection system of FIGS. 10 and 11 . 10 is a flowchart showing an example of an abnormality detection process performed for each operation period in an abnormality detection system according to a modification of the second embodiment.

Embodiments of the present disclosure will be described in detail below with reference to the drawings. Note that identical or equivalent parts in the drawings will be designated by the same reference numerals, and their descriptions will not be repeated in principle.

Embodiment 1.
1 is a block diagram showing an example of the configuration of an anomaly detection system 1 according to embodiment 1 and an air conditioning system 40 whose state is monitored by the anomaly detection system 1. As shown in FIG. 1, the anomaly detection system 1 is connected to the air conditioning system 40 via a network 900.

The anomaly detection system 1 includes a learning device 100, an inference device 200, a determination device 300, and a storage device 400. The air conditioning system 40 includes a plurality of indoor units 20, an outdoor unit 10, and a control device 30. Each of the plurality of indoor units 20 is disposed in an indoor space and connected to the outdoor unit 10. The outdoor unit 10 is disposed in a space outside the indoor space (outdoor space). The number of indoor units 20 included in the air conditioning system 40 may be one.

The outdoor unit 10 includes a compressor, an outdoor heat exchanger (first heat exchanger), and an outdoor fan. Each of the multiple indoor units 20 includes an expansion valve and an indoor heat exchanger (second heat exchanger). Refrigerant is supplied to each of the multiple indoor units 20 from the compressor included in the outdoor unit 10. The refrigerant circulates between each of the multiple indoor units 20 and the outdoor unit 10.

The control device 30 includes a thermostat and comprehensively controls the air conditioning system 40. The control device 30 is connected to the anomaly detection system 1 via a network 900. The network 900 includes the Internet and a cloud system.

Figure 2 is a functional block diagram showing the configuration of the air conditioning system 40 of Figure 1. As shown in Figure 2, the outdoor unit 10 includes a compressor 11, an outdoor heat exchanger 12 (first heat exchanger), a four-way valve 13, an outdoor fan 14, temperature sensors 51, 52, and pressure sensors 61, 62. Each of the multiple indoor units 20 includes an expansion valve 21, an indoor heat exchanger 22 (second heat exchanger), an indoor fan 23, and temperature sensors 53, 54. The temperature sensor 50 is located in the outdoor space. The expansion valve 21 includes, for example, an LEV (Linear Expansion Valve). Each of the temperature sensors 50 to 54 includes a thermistor.

The operating modes of the air conditioning system 40 include a heating mode, a cooling mode, and a defrost mode. In the heating mode, the four-way valve 13 connects the discharge port of the compressor 11 to the indoor heat exchanger 22 and also connects the outdoor heat exchanger 12 to the suction port of the compressor 11. In the heating mode, the refrigerant circulates through the compressor 11, the four-way valve 13, the indoor heat exchanger 22, the expansion valve 21, and the outdoor heat exchanger 12 in that order. In the cooling mode and the defrost mode, the four-way valve 13 connects the discharge port of the compressor 11 to the outdoor heat exchanger 12 and also connects the indoor heat exchanger 22 to the suction port of the compressor 11. In the cooling mode and the defrost mode, the refrigerant circulates through the compressor 11, the four-way valve 13, the outdoor heat exchanger 12, the expansion valve 21, and the indoor heat exchanger 22 in that order.

The temperature sensor 50 measures the temperature of the outdoor space (outdoor air temperature) and outputs the outdoor air temperature to the control device 30. The temperature sensor 51 measures the temperature of the refrigerant discharged from the compressor 11 (discharge temperature) and outputs the discharge temperature to the control device 30. The temperature sensor 52 measures the temperature (evaporation temperature or condensation temperature) of the refrigerant passing through the outdoor heat exchanger 12 and outputs the temperature to the control device 30. The temperature sensor 53 measures the temperature (condensation temperature or evaporation temperature) of the refrigerant passing through the indoor heat exchanger 22 and outputs the temperature to the control device 30. The temperature sensor 54 measures the temperature (suction temperature or blowout temperature) of the air passing through the indoor heat exchanger 22 and outputs the temperature to the control device 30. The pressure sensor 61 measures the pressure (high pressure) of the refrigerant discharged from the compressor 11 and outputs the high pressure to the control device 30. The pressure sensor 62 measures the pressure (low pressure) of the refrigerant drawn into the compressor 11 and outputs the low pressure to the control device 30.

The control device 30 controls the operating frequency of the compressor 11 to control the amount of refrigerant discharged by the compressor 11 per unit time. The control device 30 controls the opening degree of the expansion valve 21. The control device 30 controls the four-way valve 13 to switch the refrigerant circulation direction. The control device 30 controls the rotation speed of each of the outdoor fan 14 and the indoor fan 23 to control the air flow rate per unit time of the fans. The control device 30 associates operating data reflecting the state of the air conditioning system with the measurement time and transmits it to the anomaly detection system.

FIG. 3 shows an example of operational data reflecting the state of the air conditioning system 40 of FIG. 1. As shown in FIG. 3, the operational data includes, for example, outdoor air temperature, discharge temperature, evaporation temperature, condensation temperature, suction temperature, discharge temperature, high pressure, low pressure, operating frequency of the compressor 11, opening of the expansion valve 21, operating mode, operating state (operating, stopped, or standby), rotation speeds of the outdoor fan 14 and the indoor fan 23, the indoor space temperature (set temperature) set by the user, the inverter current value of the compressor 11, the inverter voltage value, the temperature of the heat sink included in the outdoor unit 10, and the temperature (liquid pipe temperature) of the liquid pipe (pipe through which liquid refrigerant flows) connecting the outdoor unit 10 and the indoor unit 20. The operating frequency of the compressor 11, opening of the expansion valve 21, and the rotation speed of the outdoor fan 14 are basic manipulated variables in VRF (Variable Refrigerant Flow) control.

The environment in which the air conditioning system 40 operates may have characteristics specific to that environment (for example, the length of the refrigerant piping, the type of indoor units 20, the number of indoor units 20, and the difference in elevation between the indoor units 20 and the outdoor unit 10). Therefore, the criteria (for example, threshold values) for detecting an abnormality in the air conditioning system 40 may differ depending on the environment in which the air conditioning system 40 operates. Therefore, if common criteria are used regardless of the environment in which the air conditioning system 40 operates, the accuracy of detecting an abnormality in the air conditioning system 40 may decrease.

Therefore, the anomaly detection system 1 generates a trained model that learns the relationship between the operating data of the air conditioning system 40 in a normal state and the normal value of the state index value (specific parameter) that represents the state of the air conditioning system 40 corresponding to that operating data. By using this trained model, it becomes possible to detect anomalies in the air conditioning system 40 using judgment criteria that are suited to the environment in which the air conditioning system 40 operates. As a result, the accuracy of anomaly detection in the air conditioning system can be improved.

Figure 4 is a block diagram showing the configuration of the anomaly detection system 1 of Figure 1. The anomaly detection system 1 acquires operating data for each of multiple consecutive operating periods (e.g., one day, one week, or one month) and constructs a trained inference model M1 using learning data including the operating data. The consecutive operating periods may be the same period (a predetermined period) or different periods. Hereinafter, the Nth operating period, which is a natural number after the start of operation of the air conditioning system 40, will be referred to as the Nth operating period, and the operating data acquired during the Nth operating period will be referred to as the Nth operating data. Figure 1 also shows the flow of data during the first operating period.

As shown in FIG. 4 , the learning device 100 includes a data acquisition unit 110 and a model generation unit 120. The inference device 200 includes a data acquisition unit 210 and an inference unit 220. The determination device 300 includes a determination unit 310 and an output unit 320. The storage device 400 stores an operating data set Dr of the air conditioning system 40 and an inference model M1. The storage device 400 may also store information necessary for calculating the state index value. The state index value includes, for example, the AK value of the outdoor heat exchanger 12 or the Cv value of the expansion valve 21. The AK value of the outdoor heat exchanger 12 represents the heat transfer performance of the outdoor heat exchanger 12 and is obtained by multiplying the heat transfer area A (m ² ) of the outdoor heat exchanger 12 by the overall heat transfer coefficient K (W/m ² ·K). The Cv value of the expansion valve 21 is a flow coefficient that is correlated with the opening degree of the expansion valve 21 and is determined by the type and opening degree of the expansion valve 21 .

The inference model M1 is a regression model including a neural network that infers the normal value of the condition index value of the air conditioning system 40 from the operating data of the air conditioning system 40. The inference model M1 may also be a classification model that infers the stage (classification) of the condition index value. The normal value of the condition index value may be the maximum and minimum values of the confidence interval that the condition index value can take when the air conditioning system 40 is normal. Furthermore, the confidence interval may be a range with the normal value as the median (for example, a range of ±10% of the normal value).

The data acquisition unit 210 of the inference device 200 acquires the first driving data, stores it in the storage device 400, and outputs it to the inference unit 220. The data acquisition unit 110 of the learning device 100 acquires the first driving data from the storage device 400 and outputs it to the model generation unit 120. Note that general AI (Artificial Intelligence) technology can be applied to clustering and weighting the parameters included in the driving data.

The model generation unit 120 performs machine learning on the inference model M1 to construct a trained inference model M1. The machine learning algorithm used by the model generation unit 120 may be a known algorithm such as supervised learning, semi-supervised learning, unsupervised learning, or reinforcement learning. Furthermore, the machine learning algorithm may be deep learning, which learns to extract features themselves, or machine learning may be performed according to other known methods, such as neural networks, genetic programming, functional logic programming, or support vector machines. Below, we will explain the case where supervised learning is applied to a neural network.

The model generation unit 120 calculates a state index value for the first operating period from the first operating data. The model generation unit 120 performs supervised learning on the inference model M1 using learning data in which the state index value for the first operating period is used as the correct answer data (teacher data) for the first operating data. The model generation unit 120 constructs a trained inference model M1 and stores it in the storage device 400. Note that since the first operating period is the period in which the air conditioning system 40 is operating for the first time at the installation location, there is a low possibility that there is an abnormality in the air conditioning system 40. Since it can be assumed that the air conditioning system 40 is normal during the first operating period, the state index value for the first operating period is used as the correct answer data for the first operating data.

Figure 5 is a block diagram showing the configuration of the anomaly detection system 1 of Figure 1 and the flow of data during the Nth operating period (N≧2). As shown in Figure 5, the data acquisition unit 210 of the inference device 200 acquires the Nth operating data and stores it in the storage device 400, and outputs the Nth operating data to the inference unit 220. The inference unit 220 uses an inference model M1 that has been trained using the first to (N-1)th operating data to infer a normal value for the state index value from the Nth operating data, and outputs the normal value and the Nth operating data to the judgment unit 310.

The determination unit 310 calculates a state index value for the Nth operating period from the Nth operating data. The determination unit 310 determines whether the air conditioning system 40 is normal during the Nth operating period by comparing the normal value from the inference unit 220 with the state index value for the Nth operating period, and outputs the determination result to the output unit 320 and the learning device 100. Note that the comparison of the normal state index value with the state index value for a certain operating period (actual state index value) includes a determination of whether the actual state index value is included in a confidence interval based on the normal value.

The model generation unit 120 associates data indicating the judgment result with the Nth driving data in the storage device 400. The output unit 320 transmits the judgment result from the judgment unit 310 to an external device (for example, a user's terminal device or the control device 30).

The data acquisition unit 110 of the learning device 100 acquires the first through Nth operating data from the storage device 400 and outputs them to the model generation unit 120. The model generation unit 120 constructs a trained inference model M1 using, as training data, the operating data acquired during the operating period when the air conditioning system 40 is operating normally, among the first through Nth operating data. The model generation unit 120 updates the inference model M1 in the storage device 400.

The anomaly detection system 1 constructs a trained inference model M1 for each predetermined operating period, thereby enabling a shorter learning period than when machine learning is performed after collecting operating data from multiple operating periods. It is possible to detect anomalies in the air conditioning system 40 at the latest after one operating period has elapsed since the start of operation of the air conditioning system 40. As a result, in addition to anomalies (wear defects or random defects) after a reference period (e.g., one year) of the air conditioning system 40 has elapsed, defects before the reference period (initial defects or random defects due to product manufacturing or installation) can be detected. Furthermore, the longer the operating period of the air conditioning system 40, the more operating data is included in the training data, and therefore the narrower the confidence interval inferred by the inference model M1. As a result, the accuracy of anomaly detection by the anomaly detection system 1 can be improved.

Figure 6 is a diagram showing an example of a neural network Nw1 included in the inference model M1 of Figures 4 and 5. As shown in Figure 6, the neural network Nw1 includes an input layer X10, an intermediate layer (hidden layer) Y10, and an output layer Z10. The input layer X10 includes neurons X11, X12, and X13. The intermediate layer Y10 includes neurons Y11 and Y12. The output layer Z10 includes neuron Z11. The input layer X10 and the intermediate layer Y10 are fully connected to each other. The intermediate layer Y10 and the output layer Z10 are fully connected to each other. The neural network Nw1 may include two or more intermediate layers.

When multiple inputs are input to neurons X11-X13 in the input layer X10, the values are multiplied by weights w11, w12, w13, w14, w15, and w16 and input to neurons Y11 and Y12 in the intermediate layer Y10. The outputs from neurons Y11 and Y12 are multiplied by weights w21 and w22 and output from neuron Z11 in the output layer Z10. The output result from output layer Z10 varies depending on the values of weights w11-w16, w21, and w22. The weights and biases of neural network Nw1 are updated by backpropagation of the error between driving data input to the input layer and the correct data so that the results output from the output layer approach the correct data.

Figure 7 shows the confidence interval SR1 based on the condition index value inferred by the trained inference model M1, together with a time chart AC of the actual condition index value. If the condition index value of the air conditioning system 40 is included in the confidence interval, the condition of the air conditioning system 40 is determined to be normal. If the condition index value of the air conditioning system 40 is not included in the confidence interval SR1, the condition of the air conditioning system 40 is determined to be abnormal. As shown in Figure 7, from time t1 onwards, the condition index value is not included in the confidence interval SR1. From time t1 onwards, the anomaly detection system 1 transmits to an external device information that an abnormality has occurred in the air conditioning system 40.

Figure 8 is a flowchart showing an example of an anomaly detection process performed for each operating period in the anomaly detection system 1 of Figures 4 and 5. Below, steps are simply referred to as S. As shown in Figure 8, in S101, the data acquisition unit 210 of the inference device 200 acquires the Nth operating data and proceeds to S102. In S102, the data acquisition unit 210 performs cleansing (preprocessing) on the Nth operating data and stores it in the storage device 400.

In S103, the inference unit 220 determines whether N is 1. If N is 1 (YES in S103), the inference unit 220 proceeds to S107. If N is not 1 (NO in S103), the inference unit 220 infers the normal value of the state index value from the Nth operating data using an inference model M1 that has been trained using, as learning data, operating data acquired during operating periods when the air conditioning system 40 is operating normally, among the first to (N-1)th operating data, and outputs the Nth operating data and the normal value to the judgment unit 310.

In S105, the judgment unit 310 calculates a state index value from the Nth driving data and outputs the judgment result as to whether the state index value is included in a confidence interval based on the normal value of the state index value to the model generation unit 120 and the output unit 320.

In S106, the model generation unit 120 determines whether the condition index value is included in the confidence interval. If the condition index value is not included in the confidence interval (NO in S106), the processing ends. If the condition index value is included in the confidence interval (YES in S106), the model generation unit 120 sets the condition of the air conditioning system 40 corresponding to the Nth operating data to normal in S107, and proceeds to S108. In S108, the model generation unit 120 constructs a learned inference model M1 by machine learning using operating data acquired during operating periods in which the air conditioning system 40 is in a normal state from among the first to Nth operating data as learning data, stores the learned inference model M1 in the storage device 400, and ends the processing.

Note that after a reference period (e.g., one year) has elapsed since the air conditioning system 40 began operating, the process of constructing the trained inference model M1 for each operating period may be stopped. Furthermore, in order to improve the accuracy of the inference model M1, it is desirable that the operating modes (e.g., heating mode, cooling mode, or defrost mode) corresponding to the multiple operating data included in the training data are the same.

Figure 9 is a block diagram showing the hardware configuration of the anomaly detection system 1 of Figure 1. As shown in Figure 9, the anomaly detection system 1 includes a processing circuit 91, a memory 92, and an input/output unit 93. The processing circuit 91 includes a CPU (Central Processing Unit) that executes programs stored in the memory 92. The processing circuit 91 may also include a GPU (Graphics Processing Unit). The functions of the anomaly detection system 1 are realized by software, firmware, or a combination of software and firmware. The software or firmware is written as a program and stored in the memory 92. The processing circuit 91 reads and executes the program stored in the memory 92. The CPU is also called a central processing unit, processing device, arithmetic unit, microprocessor, microcomputer, processor, or DSP (Digital Signal Processor).

Memory 92 includes non-volatile or volatile semiconductor memory (e.g., RAM (Random Access Memory), ROM (Read Only Memory), flash memory, EPROM (Erasable Programmable Read Only Memory), or EEPROM (Electrically Erasable Programmable Read Only Memory)), as well as magnetic disks, flexible disks, optical disks, compact disks, minidisks, or DVDs (Digital Versatile Discs). Memory 92 stores, for example, a trained inference model M1, an anomaly detection program, and a machine learning program.

The input/output unit 93 receives operations from the user and outputs processing results to the user. The input/output unit 93 includes, for example, a mouse, keyboard, touch panel, display, and speaker.

In addition, each of the learning device 100, inference device 200, and judgment device 300 shown in Figure 1 may have the hardware configuration shown in Figure 9.

In the first embodiment, the learning device 100 and the inference device 200 are described as devices separate from the air conditioning system 40 and connected to the air conditioning system 40 via the network 900, but the learning device 100 and the inference device 200 may be built into the air conditioning system 40. Furthermore, the learning device 100 and the inference device 200 may reside on a cloud server.

As described above, the anomaly detection system of embodiment 1 can improve the accuracy of anomaly detection in the air conditioning system.

Embodiment 2.
In the first embodiment, a configuration has been described in which the inference model is trained using the operating data as training data. In the second embodiment, a configuration will be described in which a simulation operating data set obtained from a simulation of the air conditioning system is added to the training data.

Figure 10 is a block diagram showing the configuration of an anomaly detection system 2 according to embodiment 2, as well as the flow of data during the first operating period. The configuration of the anomaly detection system 2 is the same as that of the anomaly detection system 1, except that a simulation operation data set Ds is added to the storage device 400 of Figure 4, and the model generation unit 120 is replaced with 120A. Other than these, the configuration of the anomaly detection system 2 is the same as that of the anomaly detection system 1.

As shown in FIG. 10, the data acquisition unit 110 acquires a simulation operation dataset Ds from the storage device 400 and outputs it to the model generation unit 120A. The simulation operation dataset Ds is operation data acquired from the results of a simulation based on parameters characterizing the air conditioning system 40 (for example, the refrigerant piping length, the type of indoor units 20, the number of indoor units 20, and the elevation difference between the indoor units 20 and the outdoor unit 10). The simulation may be performed before the start of operation of the air conditioning system 40. In the simulation, the air conditioning system 40 is assumed to be normal, and therefore the condition index value calculated by the simulation is taken as the correct data for the simulation operation dataset Ds. The model generation unit 120A constructs a learned inference model M1 using the simulation operation dataset Ds as learning data.

The data acquisition unit 210 of the inference device 200 acquires the first driving data and stores it in the storage device 400, and outputs the first driving data to the inference unit 220. The inference unit 220 uses an inference model M1 that has been trained using the simulation driving data set Ds to infer the normal value of the state index value from the first driving data, and outputs the normal value and the first driving data to the judgment unit 310.

The determination unit 310 calculates a state index value for the first operating period from the first operating data. The determination unit 310 determines whether the air conditioning system 40 is normal during the first operating period by comparing the normal value from the inference unit 220 with the state index value for the first operating period, and outputs the determination result to the output unit 320 and the learning device 100. The output unit 320 transmits the determination result from the determination unit 310 to an external device. The model generation unit 120A associates data indicating the determination result with the first operating data in the memory device 400.

Figure 11 is a block diagram showing the configuration of an anomaly detection system 2 according to embodiment 2, as well as the flow of data during the Nth operating period (N≧2). As shown in Figure 11, the data acquisition unit 210 of the inference device 200 acquires the Nth operating data and stores it in the storage device 400, and outputs the Nth operating data to the inference unit 220. The inference unit 220 uses an inference model M1 that has been trained using the first to (N-1)th operating data and the simulation operating data set Ds to infer a normal value for the state index value from the Nth operating data, and outputs the normal value and the Nth operating data to the determination unit 310. The determination unit 310 and output unit 320 perform the same processing as in embodiment 1.

The data acquisition unit 110 of the learning device 100 acquires the first to Nth driving data and the simulation driving data set Ds from the storage device 400 and outputs them to the model generation unit 120A. The model generation unit 120A performs the same processing as in embodiment 1.

The anomaly detection system 2 can construct a trained inference model M1 using a simulated driving data set Ds without using actual driving data. As a result, it is possible to achieve the same effects as in the first embodiment and to perform anomaly detection from the first driving period.

Figure 12 is a flowchart showing an example of an abnormality detection process performed for each operating period in the abnormality detection system 2 of Figures 10 and 11. The flowchart shown in Figure 12 is a flowchart in which S205 has been added to the flowchart shown in Figure 8, and S108 has been replaced with S208. Other than this, the processing in Figure 12 is the same as the processing in Figure 8, and therefore a description of the similar processing will not be repeated.

As shown in FIG. 12, steps S101 to S107 are performed in the same manner as in embodiment 1. If the answer is YES in S103, the model generation unit 120A constructs a trained inference model M1 using the simulation driving data set Ds as learning data in S205 and proceeds to S104. In S208, the model generation unit 120A constructs a trained inference model M1 using the driving data acquired during the operating period when the air conditioning system 40 was in a normal state among the first to Nth driving data and the simulation driving data set Ds as learning data, saves the trained inference model M1 in the storage device 400, and terminates the process.

Characteristics specific to the environment in which the air conditioning system 40 is operating are reflected more by the actual operating data than by the simulated operating data set Ds. Therefore, by reducing the proportion of the simulated operating data set Ds used as learning data in accordance with the increase in normal operating data included in the learning data, the accuracy of the trained inference model M1 can be improved compared to when all of the simulated operating data set Ds remain in the learning data.

Figure 13 is a flowchart showing an example of an anomaly detection process performed for each operating period in an anomaly detection system according to a modified example of embodiment 2. The flowchart shown in Figure 13 is a flowchart in which S304 and S306 have been added to the flowchart of Figure 12, and S208 has been replaced with S308. Other than this, the processing in Figure 13 is the same as in Figure 12, so a description of similar processing will not be repeated. Below, the model generation unit of the anomaly detection system according to a modified example of embodiment 2 is referred to as 120B.

As shown in FIG. 13, steps S101 to S106 are performed in the same manner as in embodiment 1. If the answer is YES in S103, the model generation unit 120B sets the usage rate of the simulation driving dataset Ds to 100% in S304, and performs S205 in the same manner as in embodiment 2. If the answer is YES in S106, the amount of normal driving data included in the learning data increases, so the model generation unit 120B reduces the usage rate of the simulation driving dataset Ds in S306, and performs S107 in the same manner as in embodiment 1. In S306, the usage rate of the simulation driving dataset Ds may be reduced by a fixed value or a fixed percentage, for example.

In S308, the model generation unit 120B constructs a learned inference model M1 using the operating data acquired during the operating period when the air conditioning system 40 is in a normal state among the first to Nth operating data and the usage percentage portion of the simulation operating data set Ds as learning data, stores it in the storage device 400, and terminates the processing.

As described above, the anomaly detection system according to embodiment 2 and the modified example can improve the accuracy of anomaly detection in the air conditioning system.

The embodiments disclosed herein are intended to be combined as appropriate within the scope of compatibility. The embodiments disclosed herein are to be considered illustrative in all respects and not restrictive. The scope of the present disclosure is defined by the claims, not the above description, and is intended to include all modifications within the meaning and scope of the claims.

1, 2 Anomaly detection system, 10 Outdoor unit, 11 Compressor, 12 Outdoor heat exchanger, 13 Four-way valve, 14 Outdoor fan, 20 Indoor unit, 21 Expansion valve, 22 Indoor heat exchanger, 23 Indoor fan, 30 Control device, 40 Air conditioning system, 50-54 Temperature sensors, 61, 62 Pressure sensors, 91 Processing circuit, 92 Memory, 93 Input/output unit, 100 Learning device, 110, 210 Data acquisition unit, 120, 120A, 120B Model generation unit, 200 Inference device, 220 Inference unit, 300 Determination device, 310 Determination unit, 320 Output unit, 400 Storage device, 900 Network, AC Time chart, Dr Operation data set, Ds Simulation operation data set, M1 Inference model, Nw1 Neural network, SR1 Confidence interval, X10 Input layer: X11 to X13, Y11, Y12, Z11 neurons, Y10 hidden layer, Z10 output layer, t1 time, w11 to w16, w21, w22 weights.

Claims (5)

A system for detecting an abnormality in an air conditioning system,
a learning device that uses learning data including operating data of the air conditioning system acquired during an operating period in which the air conditioning system is normal among first to Nth consecutive operating periods (N is a natural number) to train an inference model that infers a normal value of a specific parameter of the air conditioning system from the operating data of the air conditioning system;
an inference device that uses the inference model to infer the normal value from operating data of the air conditioning system acquired during the (N+1)th operating period;
a determination device that determines whether or not the air conditioning system is abnormal during the (N+1) operating period based on a comparison between the normal value and the specific parameter during the (N+1) operating period ,
the learning device updates the inference model using learning data including operating data of the air conditioning system acquired during an operating period in which the air conditioning system is normal among the first to (N+1) consecutive operating periods;
the inference device uses the inference model to infer the normal value from operating data of the air conditioning system acquired during the (N+2)th operating period;
The determination device determines whether or not the air conditioning system is abnormal during the (N+2) operating period based on a comparison between the normal value and the specific parameter during the (N+2) operating period . the learning data includes a simulation operation data set obtained from a simulation result of the air conditioning system;
the learning device uses the simulation driving data set to train the inference model;
the inference device infers the normal value from operating data of the air conditioning system acquired during a first operating period;
The system according to claim 1 , wherein the determination device determines whether or not the air conditioning system is abnormal during the first operating period based on a comparison between the normal value and the specific parameter during the first operating period. The system described in claim 2, wherein the learning device reduces the portion of the simulation driving data set included in the learning data as the amount of normal driving data included in the learning data increases. The air conditioning system includes an outdoor unit and at least one indoor unit,
The outdoor unit includes a compressor and a first heat exchanger,
Each of the at least one indoor unit includes an expansion valve and a second heat exchanger,
The refrigerant circulates through the compressor, the first heat exchanger, the expansion valve, and the second heat exchanger in this order, or through the compressor, the second heat exchanger, the expansion valve, and the first heat exchanger in this order,
The system according to any one of claims 1 to 3, wherein the specific parameters include at least one of an index value relating to the heat transfer performance of the first heat exchanger and a flow coefficient of the expansion valve. A method for detecting an abnormality in an air conditioning system, comprising:
constructing a trained inference model that infers normal values of specific parameters of the air conditioning system from the operating data of the air conditioning system using learning data including operating data acquired during operating periods in which the air conditioning system is normal among first to Nth consecutive operating periods (N is a natural number);
A step of inferring the normal value from operating data of the air conditioning system acquired during the (N+1)th operating period using the inference model;
determining whether or not the air conditioning system is abnormal during the (N+1) operating period based on a comparison between the normal value and the specific parameter during the (N+1) operating period ;
updating the inference model using learning data including operating data of the air conditioning system acquired during an operating period in which the air conditioning system is normal among the first to (N+1) consecutive operating periods;
Inferring the normal value from operating data of the air conditioning system acquired during the (N+2)th operating period using the inference model;
and determining whether the air conditioning system is abnormal during the (N+2) operating period based on a comparison between the normal value and the specific parameter during the (N+2) operating period .