System and method for detecting abnormality of air conditioning system

Abstract

A learning device trains an inference model to be a trained inference model using training data, the inference model being a model that infers a normal value of a specific parameter of an air conditioning system from operation data of the air conditioning system acquired in an operation period for which the air conditioning system is normal, of continuous first to N-th operation periods (N is the natural number), the training data including the operation data of the air conditioning system. An inference device, using the inference model, infers the normal value from operation data of the air conditioning system acquired in a (N+1)-th operation period. A determination device determines whether the air conditioning system in the (N+1)-th operation period is abnormal, based on comparison between the normal value and the specific parameter in the (N+1)-th operation period.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to a system and a method for detecting an abnormality of an air conditioning system.BACKGROUND ART

[0002] Conventionally, there has been known a configuration that detects an abnormality of an air conditioning system. For example, Japanese Patent Laying-Open No. 2017-221023 (PTL 1) discloses a failure sign detection device that accurately estimates an internal state of a compressor by analyzing a q-axis current that is less affected by electrical noise. According to the failure sign detection device, the accuracy of detection of an abnormality of the compressor can be improved.CITATION LISTPatent LiteraturePTL 1: Japanese Patent Laying-Open No. 2017-221023SUMMARY OF INVENTIONTechnical Problem

[0004] PTL 1 discloses that the abnormality of the compressor is detected when an intensity of an operating frequency component of the compressor exceeds a threshold value as a result of fast Fourier transform (FFT) analysis. However, the threshold value may vary depending on the operating environment of an air conditioning system. Therefore, when a common threshold value is used regardless of the operating environment of the air conditioning system, the accuracy of detection of an abnormality of the air conditioning system may decrease.

[0005] The present disclosure has been made to solve the above-described problem, and an object thereof is to improve the accuracy of detection of an abnormality of an air conditioning system.Solution to Problem

[0006] A system according to an aspect of the present disclosure detects an abnormality of an air conditioning system. The system includes: a learning device; an inference device; and a determination device. The learning device is configured to train an inference model to be a trained inference model using training data, the inference model being a model that infers a normal value of a specific parameter of the air conditioning system from operation data of the air conditioning system acquired in an operation period for which the air conditioning system is normal, of continuous first to N-th operation periods (N is the natural number), the training data including the operation data of the air conditioning system. The inference device is configured to, using the inference model, infer the normal value from operation data of the air conditioning system acquired in a (N+1)-th operation period. The determination device is configured to determine whether the air conditioning system in the (N+1)-th operation period is abnormal, based on comparison between the normal value and the specific parameter in the (N+1)-th operation period.

[0007] A method according to another aspect of the present disclosure detects an abnormality of an air conditioning system. The method includes: building a trained inference model using training data, the inference model being a model that infers a normal value of a specific parameter of the air conditioning system from operation data of the air conditioning system acquired in an operation period for which the air conditioning system is normal, of continuous first to N-th operation periods, the training data including the operation data; using the inference model, inferring the normal value from operation data of the air conditioning system acquired in a (N+1)-th operation period; and determining whether the air conditioning system in the (N+1)-th operation period is abnormal, based on comparison between the normal value and the specific parameter in the (N+1)-th operation period.Advantageous Effects of Invention

[0008] In the system and the method according to the present disclosure, the accuracy of detection of an abnormality of the air conditioning system can be improved by using the inference model trained using the training data including the operation data of the air conditioning system acquired in the operation period for which the air conditioning system is normal, of the continuous first to N-th operation periods.BRIEF DESCRIPTION OF DRAWINGS

[0009] FIG. 1 is a block diagram showing an example of configurations of an abnormality detection system according to a first embodiment and an air conditioning system whose state is monitored by the abnormality detection system.

[0010] FIG. 2 is a functional block diagram showing the configuration of the air conditioning system in FIG. 1.

[0011] FIG. 3 shows an example of operation data that reflects a state of the air conditioning system in FIG. 1.

[0012] FIG. 4 is a block diagram showing the configuration of abnormality detection system 1 in FIG. 1.

[0013] FIG. 5 is a block diagram showing together the configuration of the abnormality detection system in FIG. 1 and a data flow in an N-th operation period (N≥2).

[0014] FIG. 6 shows an example of a neural network included in an inference model in each of FIGS. 4 and 5.

[0015] FIG. 7 shows together a confidence interval based on a state index value inferred by a trained inference model and a time chart of an actual state index value.

[0016] FIG. 8 is a flowchart showing an example of an abnormality detection process performed for each operation period in the abnormality detection system in each of FIGS. 4 and 5.

[0017] FIG. 9 is a block diagram showing a hardware configuration of the abnormality detection system in FIG. 1.

[0018] FIG. 10 is a block diagram showing together a configuration of an abnormality detection system according to a second embodiment and a data flow in a first operation period.

[0019] FIG. 11 is a block diagram showing together the configuration of the abnormality detection system according to the second embodiment and a data flow in an N-th operation period (N≥2).

[0020] FIG. 12 is a flowchart showing an example of an abnormality detection process performed for each operation period in the abnormality detection system in each of FIGS. 10 and 11.

[0021] FIG. 13 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.DESCRIPTION OF EMBODIMENTS

[0022] Embodiments of the present disclosure will be described in detail hereinafter with reference to the drawings, in which the same or corresponding portions are denoted by the same reference characters and description thereof will not be repeated in principle.First Embodiment

[0023] FIG. 1 is a block diagram showing an example of configurations of an abnormality detection system 1 according to a first embodiment and an air conditioning system 40 whose state is monitored by abnormality detection system 1. As shown in FIG. 1, abnormality detection system 1 is connected to air conditioning system 40 via a network 900.

[0024] Abnormality detection system 1 includes a learning device 100, an inference device 200, a determination device 300, and a storage device 400. Air conditioning system 40 includes a plurality of indoor units 20, an outdoor unit 10 and a controller 30. Each of the plurality of indoor units 20 is arranged in an indoor space and is connected to outdoor unit 10. Outdoor unit 10 is arranged in a space (outdoor space) outside the indoor space. The number of indoor units 20 included in air conditioning system 40 may be one.

[0025] Outdoor unit 10 includes a compressor, an outdoor heat exchanger (first heat exchanger) and an outdoor fan. Each of the plurality of indoor units 20 includes an expansion valve and an indoor heat exchanger (second heat exchanger). Refrigerant is supplied from the compressor included in outdoor unit 10 to each of the plurality of indoor units 20. The refrigerant circulates between each of the plurality of indoor units 20 and outdoor unit 10.

[0026] Controller 30 includes a thermostat and controls air conditioning system 40 in an integrated manner. Controller 30 is connected to abnormality detection system 1 via network 900. Network 900 includes the Internet and a cloud system.

[0027] FIG. 2 is a functional block diagram showing the configuration of air conditioning system 40 in FIG. 1. As shown in FIG. 2, 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 and 52, and pressure sensors 61 and 62. Each of the plurality of indoor units 20 includes an expansion valve 21, an indoor heat exchanger 22 (second heat exchanger), an indoor fan 23, and temperature sensors 53 and 54. A temperature sensor 50 is arranged in the outdoor space. Expansion valve 21 includes, for example, a linear expansion valve (LEV). Each of temperature sensors 50 to 54 includes a thermistor.

[0028] Operation modes of air conditioning system 40 includes a heating mode, a cooling mode and a defrosting mode. In the heating mode, four-way valve 13 connects a discharge port of compressor 11 to indoor heat exchangers 22 and connects outdoor heat exchanger 12 to a suction port of compressor 11. In the heating mode, the refrigerant circulates in the order of compressor 11, four-way valve 13, indoor heat exchangers 22, expansion valves 21, and outdoor heat exchanger 12. In the cooling mode and the defrosting mode, four-way valve 13 connects the discharge port of compressor 11 to outdoor heat exchanger 12 and connects indoor heat exchangers 22 to the suction port of compressor 11. In the cooling mode and the defrosting mode, the refrigerant circulates in the order of compressor 11, four-way valve 13, outdoor heat exchanger 12, expansion valves 21, and indoor heat exchangers 22.

[0029] Temperature sensor 50 measures a temperature (outdoor air temperature) of the outdoor space, and outputs the outdoor air temperature to controller 30. Temperature sensor 51 measures a temperature (discharge temperature) of the refrigerant discharged from compressor 11, and outputs the discharge temperature to controller 30.

[0030] Temperature sensor 52 measures a temperature (evaporation temperature or condensation temperature) of the refrigerant passing through outdoor heat exchanger 12, and outputs the temperature to controller 30. Temperature sensor 53 measures a temperature (condensation temperature or evaporation temperature) of the refrigerant passing through indoor heat exchanger 22, and outputs the temperature to controller 30. Temperature sensor 54 measures a temperature (suction temperature or blowout temperature) of air passing through indoor heat exchanger 22, and outputs the temperature to controller 30. Pressure sensor 61 measures a pressure (high pressure) of the refrigerant discharged from compressor 11, and outputs the high pressure to controller 30. Pressure sensor 62 measures a pressure (low pressure) of the refrigerant suctioned to compressor 11, and outputs the low pressure to controller 30.

[0031] Controller 30 controls an operating frequency of compressor 11 to control an amount of the refrigerant discharged per unit time by compressor 11. Controller 30 controls a degree of opening of expansion valves 21. Controller 30 controls four-way valve 13 to switch a circulation direction of the refrigerant. Controller 30 controls a rotation speed of each of outdoor fan 14 and indoor fans 23 to control an amount of air blown per unit time by the fan. Controller 30 associates operation data that reflects the state of the air conditioning system with the measurement time, and transmits the operation data to the abnormality detection system.

[0032] FIG. 3 shows an example of the operation data that reflects the state of air conditioning system 40 in FIG. 1. As shown in FIG. 3, the operation data includes, for example, the outdoor air temperature, the discharge temperature, the evaporation temperature, the condensation temperature, the suction temperature, the blowout temperature, the high pressure, the low pressure, the operating frequency of compressor 11, the degree of opening of expansion valves 21, the operation mode, an operation state (operating, stop or standby), the rotation speed of each of outdoor fan 14 and indoor fans 23, a temperature (set temperature) of the indoor space set by a user, a current value of an inverter of compressor 11, a voltage value of the inverter, a temperature of a heat sink included in outdoor unit 10, and a temperature (liquid pipe temperature) of a liquid pipe (pipe through which liquid refrigerant flows) that connects outdoor unit 10 and indoor units 20. The operating frequency of compressor 11, the degree of opening of expansion valve 21, and the rotation speed of outdoor fan 14 are basic amounts of operation in variable refrigerant flow (VRF) control.

[0033] The operating environment of air conditioning system 40 may have characteristics (e.g., a length of a refrigerant pipe, a type of indoor units 20, the number of indoor units 20, and a height difference between indoor units 20 and outdoor unit 10) specific to the environment. Therefore, a determination criterion (e.g., threshold value) for detecting an abnormality of air conditioning system 40 may vary depending on the operating environment of air conditioning system 40. Thus, when a common determination criterion is used regardless of the operating environment of air conditioning system 40, the accuracy of detection of an abnormality of air conditioning system 40 may decrease.

[0034] Accordingly, in abnormality detection system 1, a relationship between operation data of air conditioning system 40 in a normal state and a normal value of a state index value (specific parameter) indicating a state of air conditioning system 40 and corresponding to the operation data is learned to generate a trained model. By using the trained model, an abnormality of air conditioning system 40 can be detected based on the determination criterion that matches the operating environment of air conditioning system 40. As a result, the accuracy of detection of an abnormality of the air conditioning system can be improved.

[0035] FIG. 4 is a block diagram showing the configuration of abnormality detection system 1 in FIG. 1. In abnormality detection system 1, operation data is acquired in each of a plurality of continuous operation periods (e.g., one day, one week or one month), and a trained inference model M1 is built using training data including the operation data. The plurality of continuous operation periods may be the same period as each other (predetermined period), or may be different periods from each other. In the following description, an N (natural number)-th operation period since the start of the operation of air conditioning system 40 is denoted as an N-th operation period, and operation data acquired in the N-th operation period is denoted as N-th operation data. FIG. 1 also shows a data flow in a first operation period.

[0036] As shown in FIG. 4, learning device 100 includes a data acquisition unit 110 and a model generation unit 120. Inference device 200 includes a data acquisition unit 210 and an inference unit 220. Determination device 300 includes a determination unit 310 and an output unit 320. An operation dataset Dr of air conditioning system 40 and inference model M1 are stored in storage device 400. Information required to calculate the state index value may be stored in storage device 400. The state index value includes, for example, an AK value of outdoor heat exchanger 12 or a Cv value of expansion valve 21. The AK value of outdoor heat exchanger 12 indicates the heat transfer performance of outdoor heat exchanger 12 and is obtained by a product of a heat transfer area A (m2) and a heat passage rate K (W / m2·K) of outdoor heat exchanger 12. The Cv value of expansion valve 21 is a flow rate coefficient having a correlation with the degree of opening of expansion valve 21 and is determined by the type and the degree of opening of expansion valve 21.

[0037] Inference model M1 is a regression model including a neural network, for inferring a normal value of the state index value of air conditioning system 40 from the operation data of air conditioning system 40. Inference model M1 may be a classification model that infers a stage (classification) of the state index value. The normal value of the state index value may be a maximum value and a minimum value of a confidence interval that can be taken by the state index value when air conditioning system 40 is normal. A range in which the normal value is a median value (e.g., a range that is ±10% of the normal value) may be set as the confidence interval.

[0038] Data acquisition unit 210 of inference device 200 acquires first operation data and stores the first operation data in storage device 400, and outputs the first operation data to inference unit 220. Data acquisition unit 110 of learning device 100 acquires the first operation data from storage device 400 and outputs the first operation data to model generation unit 120. A commonly used artificial intelligence (AI) technique is applicable to clustering and weighting of the parameters included in the operation data.

[0039] Model generation unit 120 performs machine learning on inference model M1 to build trained inference model M1. A machine learning algorithm used by model generation unit 120 may be a known algorithm such as supervised learning, semi-supervised learning, unsupervised learning, or reinforcement learning. Deep learning that learns extraction of a feature quantity itself can also be used as the machine learning algorithm. Machine learning may be performed in accordance with other known methods such as, for example, a neural network, genetic programming, functional logic programming, or a support vector machine. The case of applying supervised learning to the neural network will be described below.

[0040] Model generation unit 120 calculates a state index value in the first operation period from the first operation data. Model generation unit 120 performs supervised learning on inference model M1, using training data in which the state index value in the first operation period is set as ground truth data (teacher data) of the first operation data. Model generation unit 120 builds trained inference model M1 and stores trained inference model M1 in storage device 400. Since the first operation period is a time period for which air conditioning system 40 operates at the installation location for the first time, air conditioning system 40 is less likely to be abnormal. Since air conditioning system 40 in the first operation period can be assumed to be normal, the state index value in the first operation period is set as the ground truth data of the first operation data.

[0041] FIG. 5 is a block diagram showing together the configuration of abnormality detection system 1 in FIG. 1 and a data flow in the N-th operation period (N≥2). As shown in FIG. 5, data acquisition unit 210 of inference device 200 acquires the N-th operation data and stores the N-th operation data in storage device 400, and outputs the N-th operation data to inference unit 220. Inference unit 220 infers a normal value of the state index value from the N-th operation data, using inference model M1 trained using the first to (N−1)-th operation data, and outputs the normal value and the N-th operation data to determination unit 310.

[0042] Determination unit 310 calculates a state index value in the N-th operation period from the N-th operation data. Determination unit 310 determines whether air conditioning system 40 in the N-th operation period is normal, by comparing the normal value from inference unit 220 with the state index value in the N-th operation period, and outputs a result of determination to output unit 320 and learning device 100. Comparison between a normal value of a state index value and a state index value in a certain operation period (actual state index value) includes determination as to whether the actual state index value is included in a confidence interval based on the normal value.

[0043] Model generation unit 120 associates the data indicating the result of determination with the N-th operation data of storage device 400. Output unit 320 transmits the result of determination from determination unit 310 to an external device (e.g., a terminal device of the user or controller 30).

[0044] Data acquisition unit 110 of learning device 100 acquires the first to N-th operation data from storage device 400, and outputs the first to N-th operation data to model generation unit 120. Model generation unit 120 builds trained inference model M1 using, as training data, operation data acquired in an operation period for which air conditioning system 40 is normal, of the first to N-th operation data. Model generation unit 120 updates inference model M1 of storage device 400.

[0045] According to abnormality detection system 1, trained inference model M1 is built for each predetermined operation period. Therefore, a training period can be made shorter than a training period in the case of collecting operation data in a plurality of operation periods and then performing machine learning. An abnormality of air conditioning system 40 can be detected after one operation period at the latest elapses since the start of the operation of air conditioning system 40. As a result, in addition to an abnormality (a wear defect or an accidental defect) after a reference period (e.g., one year) of air conditioning system 40 elapses, a defect (an initial defect caused by product manufacturing or construction, or an accidental defect) before the reference period elapses can be detected. In addition, as the operation period of air conditioning system 40 becomes longer, the amount of the operation data included in the training data becomes larger, and thus, the confidence interval inferred by inference model M1 becomes narrower. As a result, the accuracy of detection of an abnormality by abnormality detection system 1 can be improved.

[0046] FIG. 6 shows an example of a neural network Nw1 included in inference model M1 in each of FIGS. 4 and 5. As shown in FIG. 6, neural network Nw1 includes an input layer X10, an intermediate layer (hidden layer) Y10, and an output layer Z10. Input layer X10 includes neurons X11, X12 and X13. Intermediate layer Y10 includes neurons Y11 and Y12. Output layer Z10 includes a neuron Z11. Input layer X10 and intermediate layer Y10 are fully connected to each other. Intermediate layer Y10 and output layer Z10 are fully connected to each other. Neural network Nw1 may include two or more intermediate layers.

[0047] When a plurality of inputs are input to neurons X11 to X13 of input layer X10, the values thereof are multiplied by weights w11, w12, w13, w14, w15, and w16, and are input to neurons Y11 and Y12 of intermediate layer Y10. Outputs from neurons Y11 and Y12 are multiplied by weights w21 and w22, and are output from neuron Z11 of output layer Z10. The output result from output layer Z10 varies depending on the values of weights w11 to w16 and w21 and w22. The weights and biases of neural network Nw1 are updated by back propagation with respect to an error between a result output from the output layer in response to an input of the operation data to the input layer and ground truth data such that the result approaches the ground truth data.

[0048] FIG. 7 shows together a confidence interval SR1 based on the state index value inferred by trained inference model M1 and a time chart AC of the actual state index value. When the state index value of air conditioning system 40 is included in the confidence interval, the state of air conditioning system 40 is determined as being normal. When the state index value of air conditioning system 40 is not included in confidence interval SR1, the state of air conditioning system 40 is determined as being abnormal. As shown in FIG. 7, after time t1, the state index value is not included in confidence interval SR1. The occurrence of an abnormality in air conditioning system 40 after time t1 is transmitted from abnormality detection system 1 to the external device.

[0049] FIG. 8 is a flowchart showing an example of an abnormality detection process performed for each operation period in abnormality detection system 1 in each of FIGS. 4 and 5. In the following description, the step is simply denoted as “S”. As shown in FIG. 8, in S101, data acquisition unit 210 of inference device 200 acquires the N-th operation data and moves the process to S102. In S102, data acquisition unit 210 performs cleansing (preprocessing) on the N-th operation data and stores the cleansed N-th operation data in storage device 400.

[0050] In S103, inference unit 220 determines whether N is 1. When N is 1 (YES in S103), inference unit 220 moves the process to S107. When N is not 1 (NO in S103), inference unit 220 infers the normal value of the state index value from the N-th operation data using inference model M1, and outputs the N-th operation data and the normal value to determination unit 310 in S104. Inference model M1 is a model trained using, as training data, the operation data acquired in the operation period for which air conditioning system 40 is normal, of the first to (N−1)-th operation data.

[0051] In S105, determination unit 310 calculates the state index value from the N-th operation data, and outputs, to model generation unit 120 and output unit 320, a result of determination as to whether the state index value is included in the confidence interval based on the normal value of the state index value.

[0052] In S106, model generation unit 120 determines whether the state index value is included in the confidence interval. When the state index value is not included in the confidence interval (NO in S106), model generation unit 120 ends the process. When the state index value is included in the confidence interval (YES in S106), model generation unit 120 sets the state of air conditioning system 40 corresponding to the N-th operation data to be normal in S107, and moves the process to S108. In S108, model generation unit 120 builds trained inference model M1 through machine learning in which the operation data acquired in the operation period for which the state of air conditioning system 40 is normal, of the first to N-th operation data, is used as training data, and stores trained inference model M1 in storage device 400, and then, ends the process.

[0053] After the reference period (e.g., one year) elapses from the start of the operation of air conditioning system 40, the process of building trained inference model M1 for each operation period may be stopped. In addition, in order to improve the accuracy of inference model M1, the operation modes (e.g., the heating mode, the cooling mode or the defrosting mode) corresponding to the plurality of pieces of operation data included in the training data are desirably the same as each other.

[0054] FIG. 9 is a block diagram showing a hardware configuration of abnormality detection system 1 in FIG. 1. As shown in FIG. 9, abnormality detection system 1 includes a circuitry 91, a memory 92 and an input / output unit 93. Circuitry 91 includes a central processing unit (CPU) that executes a program stored in memory 92. Circuitry 91 may include a graphics processing unit (GPU). The function of abnormality detection system 1 is implemented by software, firmware, or a combination of the software and the firmware. The software or the firmware is described as a program and stored in memory 92. Circuitry 91 reads and executes the program stored in memory 92. The CPU is also called a central processing unit, a processing unit, an operation unit, a microprocessor, a microcomputer, a processor, or a digital signal processor (DSP).

[0055] Memory 92 includes a non-volatile or volatile semiconductor memory (e.g., a random access memory (RAM), a read only memory (ROM), a flash memory, an erasable programmable read only memory (EPROM), or an electrically erasable programmable read only memory (EEPROM)), and a magnetic disk, a flexible disk, an optical disk, a compact disk, a minidisc, or a digital versatile disc (DVD). Trained inference model M1, an abnormality detection program and a machine learning program are, for example, stored in memory 92.

[0056] Input / output unit 93 receives an operation from the user and outputs a result of processing to the user. Input / output unit 93 includes, for example, a mouse, a keyboard, a touch panel, a display, and a speaker.

[0057] Each of learning device 100, inference device 200 and determination device 300 shown in FIG. 1 may have the hardware configuration shown in FIG. 9.

[0058] Although learning device 100 and inference device 200 are described in the first embodiment as the devices separate from air conditioning system 40, which are connected to air conditioning system 40 via network 900, learning device 100 and inference device 200 may built into air conditioning system 40. Alternatively, learning device 100 and inference device 200 may be present on a cloud server.

[0059] As described above, in the abnormality detection system according to the first embodiment, the accuracy of detection of an abnormality of the air conditioning system can be improved.Second Embodiment

[0060] The configuration in which the inference model is trained to be the trained inference model using the operation data as the training data is described in the first embodiment. In a second embodiment, a configuration in which a simulation operation dataset acquired from simulation of the air conditioning system is added to the training data will be described.

[0061] FIG. 10 is a block diagram showing together a configuration of an abnormality detection system 2 according to the second embodiment and a data flow in the first operation period. In the configuration of abnormality detection system 2, a simulation operation dataset Ds is added to storage device 400 in FIG. 4 and model generation unit 120 is replaced with a model generation unit 120A. The remaining configuration of abnormality detection system 2 is the same as the configuration of abnormality detection system 1.

[0062] As shown in FIG. 10, data acquisition unit 110 acquires simulation operation dataset Ds from storage device 400, and outputs simulation operation dataset Ds to model generation unit 120A. Simulation operation dataset Ds is operation data acquired from a result of simulation based on the parameters (e.g., a length of a refrigerant pipe, a type of indoor units 20, the number of indoor units 20, and a height difference between indoor units 20 and outdoor unit 10) that characterize air conditioning system40. The simulation may be performed before the start of the operation of air conditioning system 40. Since air conditioning system 40 is assumed to be normal in the simulation, a state index value calculated by the simulation is used as ground truth data of simulation operation dataset Ds. Model generation unit 120A builds trained inference model M1 using simulation operation dataset Ds as training data.

[0063] Data acquisition unit 210 of inference device 200 acquires the first operation data and stores the first operation data in storage device 400, and outputs the first operation data to inference unit 220. Inference unit 220 infers the normal value of the state index value from the first operation data, using inference model M1 trained using simulation operation dataset Ds, and outputs the normal value and the first operation data to determination unit 310.

[0064] Determination unit 310 calculates the state index value in the first operation period from the first operation data. Determination unit 310 determines whether air conditioning system 40 in the first operation period is normal, by comparing the normal value from inference unit 220 and the state index value in the first operation period, and outputs a result of determination to output unit 320 and learning device 100. Output unit 320 transmits the result of determination from determination unit 310 to the external device. Model generation unit 120A associates data indicating the result of determination with the first operation data in storage device 400.

[0065] FIG. 11 is a block diagram showing together the configuration of abnormality detection system 2 according to the second embodiment and a data flow in the N-th operation period (N≥2). As shown in FIG. 11, data acquisition unit 210 of inference device 200 acquires the N-th operation data and stores the N-th operation data in storage device 400, and outputs the N-th operation data to inference unit 220. Inference unit 220 infers the normal value of the state index value from the N-th operation data, using inference model M1 trained using the first to (N−1)-th operation data and simulation operation dataset Ds, and outputs the normal value and the N-th operation data to determination unit 310. Determination unit 310 and output unit 320 perform the same processing as that of the first embodiment.

[0066] Data acquisition unit 110 of learning device 100 acquires the first to N-th operation data and simulation operation dataset Ds from storage device 400, and outputs the first to N-th operation data and simulation operation dataset Ds to model generation unit 120A. Model generation unit 120A performs the same processing as that of the first embodiment.

[0067] According to abnormality detection system 2, trained inference model M1 can be built using simulation operation dataset Ds without using the actual operation data. As a result, the same effect as that of the first embodiment can be produced and an abnormality can be detected from the first operation period.

[0068] FIG. 12 is a flowchart showing an example of an abnormality detection process performed for each operation period in abnormality detection system 2 in each of FIGS. 10 and 11. In the flowchart shown in FIG. 12, S205 is added to the flowchart shown in FIG. 8 and S108 is replaced with S208. Since the remaining processing in FIG. 12 is the same as the processing shown in FIG. 8, description about the same processing will not be repeated.

[0069] As shown in FIG. 12, S101 to S107 are performed similarly to the first embodiment. When YES in S103, model generation unit 120A builds trained inference model M1 using simulation operation dataset Ds as training data in S205, and moves the process to S104. In S208, model generation unit 120A builds trained inference model M1 using, as training data, the operation data acquired in the operation period for which the state of air conditioning system 40 is normal, of the first to N-th operation data, and simulation operation dataset Ds, and stores trained inference model M1 in storage device 400, and then, ends the process.

[0070] The characteristics specific to the environment where air conditioning system 40 is operating are reflected more faithfully in the actual operation data than in simulation operation dataset Ds. Therefore, by decreasing a ratio of simulation operation dataset Ds used as the training data in accordance with an increase in normal operation data included in the training data, the accuracy of trained inference model M1 can be improved, as compared with the case in which the whole of simulation operation dataset Ds continues to remain in the training data.

[0071] FIG. 13 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. In the flowchart shown in FIG. 13, S304 and S306 are added to the flowchart in FIG. 12 and S208 is replaced with S308. Since the remaining processing in FIG. 13 is the same as the processing in FIG. 12, description about the same processing will not be repeated. In the following description, a model generation unit of the abnormality detection system according to the modification of the second embodiment will be referred to as 120B.

[0072] As shown in FIG. 13, S101 to S106 are performed similarly to the first embodiment. When YES in S103, model generation unit 120B sets a use ratio of simulation operation dataset Ds to 100% in S304, and performs S205 similarly to the second embodiment. When YES in S106, the normal operation data included in the training data increases. Therefore, model generation unit 120B decreases the use ratio of simulation operation dataset Ds in S306, and performs S107 similarly to the first embodiment. In S306, the use ratio of simulation operation dataset Ds may be decreased by a certain value, or may be decreased by a certain ratio, for example.

[0073] In S308, model generation unit 120B builds trained inference model M1 using, as training data, the operation data acquired in the operation period for which the state of air conditioning system 40 is normal, of the first to N-th operation data, and the use ratio portion of simulation operation dataset Ds, and stores trained inference model M1 in storage device 400, and then, ends the process.

[0074] As described above, in the abnormality detection system according to the second embodiment and the modification, the accuracy of detection of an abnormality of the air conditioning system can be improved.

[0075] The embodiments disclosed herein are also planned to be implemented in appropriate combination within a non-contradictory range. It should be understood that the embodiments disclosed herein are illustrative and non-restrictive in every respect. The scope of the present disclosure is defined by the terms of the claims, rather than the description above, and is intended to include any modifications within the scope and meaning equivalent to the terms of the claims.REFERENCE SIGNS LIST1, 2 abnormality 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 controller; 40 air conditioning system; 50 to 54 temperature sensor; 61, 62 pressure sensor; 91 circuitry; 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 dataset; Ds simulation operation dataset; M1 inference model; Nw1 neural network; SR1 confidence interval; X10 input layer; X11 to X13, Y11, Y12, Z11 neuron; Y10 intermediate layer; Z10 output layer; t1 time; w11 to w16, w21, w22 weight.

Claims

1. A system for detecting an abnormality of an air conditioning system, the system comprising:a learning device configured to train an inference model to be a trained inference model using training data, the inference model being a model that infers a normal value of a specific parameter of the air conditioning system from operation data of the air conditioning system acquired in an operation period for which the air conditioning system is normal, of continuous first to N-th operation periods (N is the natural number), the training data including the operation data of the air conditioning system;an inference device configured to, using the inference model, infer the normal value from operation data of the air conditioning system acquired in a (N+1)-th operation period; anda determination device configured to determine whether the air conditioning system in the (N+1)-th operation period is abnormal, based on comparison between the normal value and the specific parameter in the (N+1)-th operation period, whereinthe learning device is configured to update the inference model using training data including operation data of the air conditioning system acquired in an operation period for which the air conditioning system is normal, of the continuous first to (N+1)-th operation periods,the inference device is configured to, using the inference model, infer the normal value from operation data of the air conditioning system acquired in a (N+2)-th operation period, andthe determination device is configured to determine whether the air conditioning system in the (N+2)-th operation period is abnormal, based on comparison between the normal value and the specific parameter in the (N+2)-th operation period.

2. The system according to claim 1, whereinthe training data includes a simulation operation dataset acquired from a simulation result of the air conditioning system,the learning device is configured to train the inference model to be the trained inference model using the simulation operation dataset,the inference device is configured to infer the normal value from operation data of the air conditioning system acquired in a first operation period, andthe determination device is configured to determine whether the air conditioning system in the first operation period is abnormal, based on comparison between the normal value and the specific parameter in the first operation period.

3. The system according to claim 2, whereinthe learning device is configured to decrease a portion of the simulation operation dataset included in the training data, as normal operation data included in the training data increases.

4. The system according to claim 1, whereinthe 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,refrigerant circulates in order of the compressor, the first heat exchanger, the expansion valve, and the second heat exchanger, or circulates in order of the compressor, the second heat exchanger, the expansion valve, and the first heat exchanger, andthe specific parameter includes at least one of an index value relating to heat transfer performance of the first heat exchanger and a flow rate coefficient of the expansion valve.

5. A method for detecting an abnormality of an air conditioning system, the method comprising:building a trained inference model using training data, the inference model being a model that infers a normal value of a specific parameter of the air conditioning system from operation data of the air conditioning system acquired in an operation period for which the air conditioning system is normal, of continuous first to N-th operation periods (N is the natural number), the training data including the operation data;using the inference model, inferring the normal value from operation data of the air conditioning system acquired in a (N+1)-th operation period;determining whether the air conditioning system in the (N+1)-th operation period is abnormal, based on comparison between the normal value and the specific parameter in the (N+1)-th operation period;updating the inference model using training data including operation data of the air conditioning system acquired in an operation period for which the air conditioning system is normal, of the continuous first to (N+1)-th operation periods;using the inference model, inferring the normal value from operation data of the air conditioning system acquired in a (N+2)-th operation period; anddetermining whether the air conditioning system in the (N+2)-th operation period is abnormal, based on comparison between the normal value and the specific parameter in the (N+2)-th operation period.

Images