Learning Device and Inference Device for Maintenance of Air Conditioning Equipment

Abstract

The model generation unit (120) sets the first model (M1), the second model (M2), and the third model (M3) as learned models respectively. The first learning data (Ld1) includes: a first parameter (Prm1) indicating the degree of clogging of the air filter, a second parameter (Prm2) related to the air conditioning intensity of the air conditioning system, and a third parameter (Prm3) indicating the increase in the power cost of the air conditioning system caused by the first parameter (Prm1) in the case of operating with the second parameter (Prm2). The second learning data (Ld2) includes: a fourth parameter (Prm4) indicating the first date and time, and a fifth parameter (Prm5) related to the air conditioning intensity of the air conditioning system assumed at the first date and time. The third learning data (Ld3) includes: a sixth parameter (Prm6) indicating the second date and time, and a seventh parameter (Prm7) indicating the maintenance cost of the air filter at the second date and time.

Description

Technical Field

[0001] The present disclosure relates to a learning device and an inference device for the maintenance of air conditioning equipment. Background Art

[0002] Conventionally, a device for detecting the maintenance timing of an air filter for an air conditioner has been known. For example, Japanese Patent Laid-Open No. 7-63405 (Patent Document 1) discloses a dirt display device for an air filter. The dirt display device for an air filter detects a change in the fan air volume caused by the clogging of the air filter as a change in the torque of the fan motor in a convection type air conditioner that controls the rotational speed of the fan motor. The dirt display device for an air filter uses the fact that the torque corresponds to the current value supplied to the fan motor to grasp the clogging state of the air filter. According to the dirt display device for an air filter, by reliably detecting the clogging state of the air filter and reporting it to the outside, it is possible to promote the cleaning or replacement of the air filter, and thus it is possible to prevent the reduction of the function of the fan motor, the increase in noise, and mechanical failures.

[0003] Patent Document 1: Japanese Patent Laid-Open No. 7-63405

[0004] Even if it is possible to detect the clogging of the air filter that causes a decrease in the power efficiency of the air conditioning device, the maintenance (for example, cleaning and replacement) of the air filter requires cost. In addition, the cost required for the maintenance of the air filter varies depending on the type of the air conditioner (for example, ceiling type or duct type), the number of air filters, the difference in the clogging degree caused by the installation state of the air filter, the labor cost, and the implementation timing of the cleaning and replacement of the air filter. That is, in order to determine the optimal timing for air filter maintenance, it is necessary to comprehensively consider multiple factors. However, the method for detecting the clogging of the air filter disclosed in Patent Document 1 is a rule-based method based on the comparison of the current value with a threshold value, and thus it is impossible to reflect factors other than the current value in the determination of the maintenance timing of the air filter. Summary of the Invention

[0005] The present disclosure has been made to solve the above-described problems, and an object thereof is to reduce the maintenance cost of the air filter.

[0006] A learning device according to one aspect of the present disclosure learns the maintenance of an air conditioning system including at least one air filter. The learning device includes a first data acquisition unit and a model generation unit. The first data acquisition unit acquires first learning data, second learning data, and third learning data. The model generation unit uses the first learning data, the second learning data, and the third learning data to set the first model, the second model, and the third model as learned models respectively. The first learning data includes: a first parameter indicating the degree of clogging of at least one air filter, a second parameter related to the air conditioning intensity of the air conditioning system, and a third parameter indicating the increase in the power cost of the air conditioning system caused by the first parameter when operating with the second parameter. The second learning data includes: a fourth parameter indicating a first date and time, and a fifth parameter related to the air conditioning intensity of the air conditioning system assumed at the first date and time. The third learning data includes: a sixth parameter indicating a second date and time, and a seventh parameter indicating the maintenance cost of at least one air filter at the second date and time. The first model estimates the third parameter based on the first parameter and the second parameter. The second model estimates the fifth parameter based on the fourth parameter. The third model estimates the seventh parameter based on the sixth parameter.

[0007] An inference device according to another aspect of the present disclosure uses the learned first model, the learned second model, and the learned third model to infer the maintenance of an air conditioning system including at least one air filter. The first model estimates the third parameter based on the first parameter and the second parameter. The second model estimates the fifth parameter based on the fourth parameter. The third model estimates the seventh parameter based on the sixth parameter. The first parameter indicates the degree of clogging of at least one air filter. The second parameter is a parameter related to the air conditioning intensity of the air conditioning system. The third parameter indicates the increase in the power cost of the air conditioning system caused by the first parameter when operating with the second parameter. The fourth parameter indicates a first date and time. The fifth parameter indicates the air conditioning intensity of the air conditioning system assumed at the first date and time. The sixth parameter indicates a second date and time. The seventh parameter indicates the maintenance cost of at least one air filter at the second date and time. The inference device includes a data acquisition unit and an inference unit. The data acquisition unit acquires the first parameter, the second parameter, the fourth parameter, and the fifth parameter. The inference unit uses the first model, the second model, and the third model. The inference unit uses the first model and estimates the third parameter based on the first parameter and the second parameter, uses the second model and estimates the fifth parameter based on the fourth parameter, and uses the third model and estimates the seventh parameter based on the sixth parameter.

[0008] According to the learning device and the inference device of the present disclosure, the first model estimates the third parameter based on the first parameter and the second parameter, the second model estimates the fifth parameter based on the fourth parameter, and the third model estimates the seventh parameter based on the sixth parameter, thereby being able to reduce the maintenance cost of the air filter. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 It is a block diagram showing an example of the structure of an air filter maintenance system including a learning device and an inference device according to an embodiment, and an air conditioning system that monitors the maintenance timing of an air filter through the air filter maintenance system.

[0010] Figure 2 It is used to illustrate Figure 1 the air flow of each indoor unit of

[0011] Figure 3 It is a diagram showing Figure 1 a curve showing an example of the annual change in the power cost of the air conditioning system of Figure 2 a curve showing an example of the annual change in the cost required for the maintenance of the air filter of

[0012] Figure 4 It is a block diagram showing Figure 1 the structure of the learning device of

[0013] Figure 5 It is a diagram showing an example of a neural network.

[0014] Figure 6 It is a flowchart showing Figure 4 the learning process of the learning device of

[0015] Figure 7 It is a block diagram showing Figure 1 the structures of the inference device and the determination device of

[0016] Figure 8 It is a block diagram showing Figure 7 the inference process of the inference device of

[0017] Figure 9 It is a block diagram showing Figure 1 the hardware structure of the air filter maintenance system of Detailed Embodiment

[0018] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. In addition, the same or corresponding parts in the drawings are denoted by the same reference numerals and their descriptions will not be repeated in principle.

[0019] Figure 1 It is a block diagram showing an example of the structure of an air filter maintenance system 10 including a learning device 100 and an inference device 200 according to an embodiment, and an air conditioning system 30 that monitors the maintenance timing of an air filter through the air filter maintenance system 10. As Figure 1As shown, the air filter maintenance system 10 includes a learning device 100, an inference device 200, and a determination device 300. The air conditioning system 30 includes indoor units 31A and 31B, an outdoor unit 32, and a control device 33. The indoor units 31A and 31B are connected to the outdoor unit 32. The outdoor unit 32 includes a compressor, an outdoor heat exchanger, an expansion valve, and a fan. Refrigerant is supplied from the compressor to the indoor units 31A and 31B respectively. The refrigerant circulates between the indoor unit 31A and the outdoor unit 32, and also circulates between the indoor unit 31B and the outdoor unit 32. The control device 33 includes a thermostat and integrally controls the air conditioning system 30. The control device 33 is connected to the air filter maintenance system 10 via a network 900. The network 900 includes the Internet and a cloud system.

[0020] Figure 2 is for explaining Figure 1 the air flow of each of the indoor units 31A and 31B. As Figure 2 shown, the indoor units 31A and 31B each include an indoor heat exchanger 311, a fan 312, and an air filter 313. By the air flow formed by the fan 312, return air RA (Return Air) is inhaled from the outside of each of the indoor units 31A and 31B. The return air RA passes through the air filter 313 and heads towards the indoor heat exchanger 311. The air heated or cooled by the indoor heat exchanger is supplied as supply air SA (Supply Air) to the air-conditioned space. Dust and the like contained in the return air RA are removed by the air filter 313. If the usage period of the air filter 313 becomes longer, dust and the like accumulate and clog the air filter 313, the dust-proof effect of the air filter 313 decreases, and the amount of air passing through the air filter 313 per unit time decreases. In order to maintain the amount of air per unit time supplied to the air-conditioned space, the lower the amount of air passing through the air filter 313 per unit time, the more it is necessary to increase the rotational speed of the fan 312. That is, in order to maintain the air conditioning intensity in a state where the air filter is clogged, it is necessary to increase the power consumption used in the air conditioning system 30. In order to suppress the increase in power costs, it is necessary to maintain (clean or replace) the air filter 313 at an appropriate timing.

[0021] Figure 3 is a curve Cp1, Cp2, Cp3 showing an example of the annual change in the power cost of the Figure 1 air conditioning system 30 representing Figure 2A graph showing together the curve Cm of an example of the annual change in the cost required for maintenance of the air filter 313 and the curve Tmp of an example of the annual change in the temperature. The curve Cp1 represents the change in the power cost when the air filter 313 is used for one year from new. The curve Cp2 represents the power cost when the air filter 313 that has been used for one year is used for one more year. The curve Cp3 represents the power cost when the air filter 313 that has been used for two years is used for one more year. As Figure 3 shown, since the clogging of the air filter 313 becomes more severe as the usage period of the air filter 313 becomes longer, the power cost increases in the order of the curves Cp1 to Cp3 during the vast majority of the year. In particular, during the period from January to March when the temperature drops the most and during the period from June to September when the temperature rises the most, the increase in the power cost corresponding to the usage period is significant. For example, the increase in the power cost from January 1 of the second year after the new air filter 313 is used until the date and time dt1 is the value obtained by integrating the absolute value of the difference between the curves Cp2 and Cp1 from January 1 to the date and time dt1 (the area of the region Rg1). Therefore, the necessity of maintaining the air filter is high during the period from January to March and from June to September. However, the maintenance cost of the air filter is the highest during the period from January to March and from September to December. Thus, in order to determine the optimal maintenance timing of the air filter, various factors need to be comprehensively considered. Therefore, it is difficult to determine the optimal maintenance timing of the air filter based on a rule-based method using a unified judgment criterion.

[0022] Therefore, in the air filter maintenance system 10, learned models that have respectively learned the relationships between the clogging degree, the air conditioner intensity, and the cost increase amount, the relationship between the date and time and the air conditioner intensity, and the relationship between the date and time and the maintenance cost are generated. By using these learned models, the clogging degree, the air conditioner intensity, the cost increase amount, the date and time, and the maintenance cost can be comprehensively considered. By comprehensively considering, the optimal maintenance timing of the air filter can be determined, and thus the maintenance cost of the air filter can be reduced.

[0023] Figure 4 is a block diagram showing the structure of the learning device 100 that Figure 1 learns. As Figure 4 shown, the learning device 100 includes a data acquisition unit 110 (first data acquisition unit) and a model generation unit 120. An increased cost estimation model M1, an air conditioner intensity estimation model M2, and a maintenance cost estimation model M3 are stored in a learned model storage unit 140 provided outside the learning device 100. In addition, the learned model storage unit 140 may be formed inside the learning device 100.

[0024] The increased cost estimation model M1 receives the clogging degree parameter Prm1 (the first parameter) and the air conditioner intensity control parameter Prm2 (the second parameter), and outputs the increased cost parameter Prm3 (the third parameter). The air conditioner intensity estimation model M2 receives the date and time parameter Prm4 (the fourth parameter) and outputs the air conditioner intensity control parameter Prm5 (the fifth parameter). The maintenance cost estimation model M3 receives the date and time parameter Prm6 (the sixth parameter) and outputs the maintenance cost parameter Prm7 (the seventh parameter). The increased cost estimation model M1, the air conditioner intensity estimation model M2, and the maintenance cost estimation model M3 each include a neural network.

[0025] The data acquisition unit 110 acquires the clogging degree parameter Prm1, the air conditioner intensity control parameter Prm2, and the increased cost parameter Prm3 as the learning data Ld1. The clogging degree parameter Prm1 represents the degree of clogging of the air filter 313 from 0% to 100%. The air conditioner intensity control parameter Prm2 includes the on / off of the thermostat, the rotation frequency of the compressor, the wind force of the fan, the evaporation temperature of the refrigerant, and the condensation temperature of the refrigerant. The increased cost parameter Prm3 is the correct answer data indicating the increase in the power cost generated in the clogging degree parameter Prm1 and the air conditioner intensity control parameter Prm2, and is the increased amount of the power cost based on the case where the clogging degree parameter of the air filter 313 is 0%. For example, when the clogging degree parameter Prm1 is 60%, the air conditioner intensity corresponding to the air conditioner intensity control parameter Prm2 is level X, and the increased cost parameter Prm3 is the increased amount of the power cost per hour, the increased cost parameter Prm3 is 1.5 yen.

[0026] The model generation unit 120 uses the learning data Ld1 to learn the relationship between the clogging degree parameter Prm1 and the air conditioner intensity control parameter Prm2 and the increased cost parameter Prm3. The learning data Ld1 is created using a combination of the clogging degree parameter Prm1, the air conditioner intensity control parameter Prm2, and the increased cost parameter Prm3. The model generation unit 120 sets the increased cost estimation model M1 as a learned model using the learning data Ld1.

[0027] The data acquisition unit 110 acquires the date and time parameter Prm4 and the air conditioner intensity control parameter Prm5 as the learning data Ld2. The air conditioner intensity control parameter Prm5 is the correct answer data indicating the air conditioner intensity assumed at the date and time determined by the date and time parameter Prm4. The air conditioner intensity control parameter Prm5 includes the on / off of the thermostat, the rotation frequency of the compressor, the wind force of the fan, the evaporation temperature of the refrigerant, and the condensation temperature of the refrigerant.

[0028] The model generation unit 120 uses the learning data Ld2, which is created using the combination of the date and time parameter Prm4 and the air conditioner intensity control parameter Prm5, to learn the relationship between the date and time parameter Prm4 and the air conditioner intensity control parameter Prm5. The model generation unit 120 sets the air conditioner intensity estimation model M2 as a learned model using the learning data Ld2.

[0029] The data acquisition unit 110 acquires the date and time parameter Prm6 and the maintenance cost parameter Prm7 as the learning data Ld3. The maintenance cost parameter Prm7 is the correct answer data indicating the maintenance cost of the air filter required at the date and time determined by the date and time parameter Prm6. The maintenance cost parameter Prm7 is, for example, the amount per day or the amount per hour.

[0030] The model generation unit 120 uses the learning data Ld3, which is created using the combination of the date and time parameter Prm6 and the maintenance cost parameter Prm7, to learn the relationship between the date and time parameter Prm6 and the maintenance cost parameter Prm7. The model generation unit 120 sets the maintenance cost estimation model M3 as a learned model using the learning data Ld3.

[0031] The learning algorithm used by the model generation unit 120 can also be a well-known algorithm such as supervised learning, unsupervised learning, or reinforcement learning. Hereinafter, the case of applying a neural network will be described.

[0032] The model generation unit 120 learns the cost increase amount, the air conditioner intensity, and the maintenance cost of the air filter, for example, through a neural network model, so-called supervised learning. Here, supervised learning refers to a method of learning the features included in these learning data by giving a set of input and result (label) data to the learning device 100 and inferring the result based on the input.

[0033] A neural network is composed of the following layers: an input layer composed of multiple neurons, an intermediate layer (hidden layer) composed of multiple neurons, and an output layer composed of multiple neurons. The intermediate layer can also be one layer or two or more layers.

[0034] Figure 5 is a diagram showing a neural network Nw1 as an example of a neural network. As Figure 5 shown, the neural network Nw1 includes an input layer X10, an intermediate 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 neurons Z11, Z12, and Z13. 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.

[0035] When multiple inputs are respectively input into the neurons X11 to X13 of the input layer X10, the above values are multiplied by the weights w11 to w16 and input into the neurons Y11 and Y12 of the intermediate layer Y10. The outputs from the neurons Y11 and Y12 are multiplied by the weights w21 to w26 and output from the neurons Z11 to Z13 of the output layer Z10. The output result from the output layer Z10 changes according to the values of the weights w11 to w16 and w21 to w26.

[0036] The neural network of the increased cost estimation model M1 learns the increased cost through supervised learning based on the learning data Ld1 created using the combination of the clogging degree, air-conditioning intensity, and increased cost (correct answer data) obtained by the data acquisition unit 110. That is, the weights and biases of the neural network of the increased cost estimation model M1 are updated by backpropagation of the error between this result and the correct answer data, so that the result output from the output layer when the clogging degree and air-conditioning intensity are input to the input layer approaches the increased cost (correct answer data).

[0037] The neural network of the air-conditioning intensity estimation model M2 learns the air-conditioning intensity through supervised learning based on the learning data Ld2 created based on the combination of the date and time and the air-conditioning intensity (correct answer data). That is, the weights and biases of the neural network of the air-conditioning intensity estimation model M2 are updated by backpropagation of the error between this result and the correct answer data, so that the result output from the output layer when the date and time are input to the input layer approaches the air-conditioning intensity (correct answer data).

[0038] The maintenance cost estimation model M3 learns the maintenance cost of the air filter through supervised learning based on the learning data Ld3 created using the combination of the date and time and the maintenance cost (correct answer data). That is, the weights and biases of the neural network of the maintenance cost estimation model M3 are updated by backpropagation of the error between this result and the correct answer data, so that the result output from the output layer when the date and time are input to the input layer approaches the maintenance cost of the air filter (correct answer data).

[0039] Figure 6 represents Figure 4 the flowchart of the learning process of the learning device 100. Hereinafter, the steps will be abbreviated as S. As Figure 6 shown, in S101, the data acquisition unit 110 acquires the learning data Ld1 to Ld3. In addition, the clogging degree parameter Prm1, the air-conditioning intensity control parameter Prm2, and the increased cost parameter Prm3 do not need to be acquired simultaneously as long as they are associated, and can also be acquired at different times. The same applies to the date and time parameter Prm4 and the air-conditioning intensity control parameter Prm5, and the date and time parameter Prm6 and the maintenance cost parameter Prm7.

[0040] In S102, based on the learning data Ld1 to Ld3 obtained by the data acquisition unit 110, the model generation unit 120 learns the cost increase amount, the air conditioner intensity, and the maintenance cost of the air filter through so-called supervised learning, and sets the increased cost estimation model M1, the air conditioner intensity estimation model M2, and the maintenance cost estimation model M3 as the learned models respectively.

[0041] In S103, the model generation unit 120 stores the increased cost estimation model M1 of the learned model, the air conditioner intensity estimation model M2 of the learned model, and the maintenance cost estimation model M3 of the learned model in the learned model storage unit 140.

[0042] Figure 7 It represents Figure 1 The block diagram of the structure of the inference device 200 and the determination device 300. The inference device 200 includes a data acquisition unit 210 and an inference unit 220. The determination device 300 includes an integration unit 310, a determination unit 320, and an output unit 330.

[0043] The data acquisition unit 210 acquires the clogging degree parameter Prm1, the air conditioner intensity control parameter Prm2, the date and time parameter Prm4, and the date and time parameter Prm6. The conventional method is used for the detection of the clogging degree parameter Prm1 of the air filter. The inference unit 220 uses the learned models M1 to M3 stored in the learned model storage unit 140 to estimate the cost increase amount parameter Prm3, the air conditioner intensity control parameter Prm5, and the maintenance cost parameter Prm7. That is, by inputting the clogging degree parameter Prm1, the air conditioner intensity control parameter Prm2, the date and time parameter Prm4, and the date and time parameter Prm6 acquired by the data acquisition unit 210 into the learned models M1 to M3, the cost increase amount parameter Prm3, the air conditioner intensity control parameter Prm5, and the maintenance cost parameter Prm7 can be estimated. In addition, in the embodiment, the structure of using the learned models learned by the model generation unit 120 in Figure 3 to estimate the cost increase amount parameter Prm3, the air conditioner intensity control parameter Prm5, and the maintenance cost parameter Prm7 has been described, but learned models learned in other environments can also be used to output the cost increase amount parameter Prm3, the air conditioner intensity control parameter Prm5, and the maintenance cost parameter Prm7.

[0044] Figure 8 It represents Figure 7 The flowchart of the inference process of the inference device 200. As Figure 8As shown, in S201, the data acquisition unit 210 acquires the clogging degree parameter Prm1, the air conditioner intensity control parameter Prm2, the date and time parameter Prm4, and the date and time parameter Prm6. In S202, the inference unit 220 inputs the clogging degree parameter Prm1, the air conditioner intensity control parameter Prm2, the date and time parameter Prm4, and the date and time parameter Prm6 into the learned models M1 to M3 stored in the learned model storage unit 140, and acquires the cost increase amount parameter Prm3, the air conditioner intensity control parameter Prm5, and the maintenance cost parameter Prm7. In S203, the integration unit 310 integrates the cost increase amount parameter Prm3, the air conditioner intensity control parameter Prm5, and the maintenance cost parameter Prm7 obtained from the learned models M1 to M3. In S204, the determination unit 320 calculates the comprehensive cost when maintaining the air filter 313 using the cost increase amount parameter Prm3 output from the learned cost increase estimation model M1, the air conditioner intensity corresponding to the air conditioner intensity control parameter Prm5 output from the learned air conditioner intensity estimation model M2, and the maintenance cost of the air filter 313 output from the maintenance cost estimation model M3, and determines the timing when this comprehensive cost becomes the minimum. In S205, the output unit 330 outputs this timing to the control device 33 of the air conditioning system 30.

[0045] As described above, according to the inference device 200 and the determination device 300, by estimating the cost increase caused by important factors such as the reduction in the efficiency of the air conditioning equipment due to the clogging of the air filter with respect to the installed model of the air filter, the installation location in space, and the air conditioner intensity over time, and the maintenance cost of the air filter that varies depending on the implementation period, it is possible to output the optimal maintenance timing of the air filter.

[0046] In addition, in the embodiment, the case where supervised learning is applied to the learning algorithm used by the model generation unit 120 has been described, but the learning algorithm is not limited to supervised learning. In addition to supervised learning, reinforcement learning, unsupervised learning, or semi-supervised learning can also be applied.

[0047] In addition, the model generation unit 120 can also use the learning data obtained from multiple air conditioning systems 30 to learn the cost increase amount, the air conditioning intensity, and the maintenance cost of the air filter. In addition, the model generation unit 120 can obtain learning data from multiple air conditioning systems 30 used in the same area, or can use the learning data collected from multiple air conditioning systems 30 that operate independently in different areas to learn the cost increase amount, the air conditioning intensity, and the maintenance cost of the air filter. Additionally, the air conditioning system 30 from which the learning data is collected can be added to the learning object midway or removed from the learning object. Also, the learning device 100 that has learned the cost increase amount, the air conditioning intensity, and the maintenance cost of the air filter for a certain air conditioning system 30 can be applied to other air conditioning systems 30, and the cost increase amount, the air conditioning intensity, and the maintenance cost of the air filter can be learned and updated again for that other air conditioning system 30.

[0048] In addition, as the learning algorithm used in the model generation unit 120, deep learning that extracts the learning feature amount itself can also be used, and mechanical learning can be performed according to other well-known methods, such as neural networks, genetic programming, functional logic programming, or support vector machines.

[0049] In addition, in the embodiment, although it is described that the learning device 100 and the inference device 200 are connected to the air conditioning system 30 via the network 900 and are devices different from the air conditioning system 30, the learning device 100 and the inference device 200 can also be built into the air conditioning system 30. Additionally, the learning device 100 and the inference device 200 can also exist on a cloud server.

[0050] Figure 9 is a block diagram showing Figure 1 the hardware structure of the air filter maintenance system 10. As Figure 9 shown, the air filter maintenance system 10 includes a processing circuit 51, a memory 52 (storage unit), and an input / output unit 53. The processing circuit 51 includes a CPU (Central Processing Unit) that executes the program stored in the memory 52. The processing circuit 51 can also include a GPU (Graphics Processing Unit). The functions of the air filter maintenance system 10 are realized by software, firmware, or a combination of software and firmware. The software or firmware is described as a program and stored in the memory 52. The processing circuit 51 reads and executes the program stored in the memory 52. In addition, the CPU is also referred to as a central processing device, a processing device, an arithmetic device, a microprocessor, a microcomputer, a processor, or a DSP (Digital Signal Processor).

[0051] The memory 52 includes non-volatile or volatile semiconductor memories (such as 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, floppy disks, optical disks, compact discs, mini discs, or DVDs (Digital Versatile Discs). Stored in the memory 52 are, for example, a learned model, an air filter maintenance program, and a machine learning program.

[0052] The input / output unit 53 receives operations from the user and outputs the processing results to the user. The input / output unit 53 includes, for example, a mouse, a keyboard, a touch panel, a display, and a speaker.

[0053] As described above, the learning device and the inference device according to the embodiment can reduce the maintenance cost of the air filter.

[0054] The embodiments disclosed herein should be considered illustrative in all respects and not restrictive. The scope of the present disclosure is not defined by the above description but by the claims, and is intended to include all modifications within the meaning and scope equivalent to the claims.

[0055] Description of Reference Numerals

[0056] 10... Air filter maintenance system; 30... Air conditioning system; 31A, 31B... Indoor units; 32... Outdoor unit; 33... Control device; 51... Processing circuit; 52... Memory; 53... Input / output unit; 100... Learning device; 110, 210... Data acquisition units; 120... Model generation unit; 140... Learned model storage unit; 200... Inference device; 220... Inference unit; 300... Determination device; 310... Integration unit; 311... Indoor heat exchanger; 312... Fan; 313... Air filter; 320... Determination unit; 330... Output unit; 900... Network; Ld1, Ld2, Ld3... Learning data; M1... Increased cost estimation model; M2... Air conditioning intensity estimation model; M3... Maintenance cost estimation model; Nw1... Neural network; Prm1... Blockage degree parameter; Prm2, Prm5... Air conditioning intensity control parameters; Prm3... Cost increase amount parameter; Prm4, Prm6... Date and time parameters; Prm7... Maintenance cost parameter.

Claims

1. A learning device for learning the maintenance of an air conditioning system including at least one air filter, characterized in that, Comprising: A first data acquisition unit that acquires first learning data, second learning data, and third learning data; and A model generation unit that uses the first learning data, the second learning data, and the third learning data to set a first model, a second model, and a third model as learned models respectively, The first learning data includes: a first parameter indicating the degree of clogging of the at least one air filter, a second parameter related to the air conditioning intensity of the air conditioning system, and a third parameter indicating the increase in the power cost of the air conditioning system caused by the first parameter when operating with the second parameter, The second learning data includes: a fourth parameter indicating a first date and time, and a fifth parameter related to the air conditioning intensity of the air conditioning system assumed at the first date and time, The third learning data includes: a sixth parameter indicating a second date and time, and a seventh parameter indicating the maintenance cost of the at least one air filter at the second date and time, The first model estimates the third parameter based on the first parameter and the second parameter, The second model estimates the fifth parameter based on the fourth parameter, The third model estimates the seventh parameter based on the sixth parameter.

2. The learning device according to claim 1, wherein The third parameter, the fifth parameter, and the seventh parameter acquired by the first data acquisition unit are correct answer data respectively, The model generation unit performs supervised learning on the first model, the second model, and the third model respectively.

3. An inference device, characterized in that, Comprising: A second data acquisition unit that acquires a first parameter indicating the degree of clogging of at least one air filter, a second parameter related to the air conditioning intensity of the air conditioning system, a fourth parameter indicating a first date and time, and a sixth parameter indicating a second date and time; And An inference unit that uses the learned first model, the learned second model, and the learned third model generated by the learning device according to claim 1 or 2, The inference unit uses the learned first model and estimates the third parameter based on the first parameter and the second parameter acquired by the second data acquisition unit, uses the learned second model and estimates the fifth parameter based on the fourth parameter acquired by the second data acquisition unit, and uses the learned third model and estimates the seventh parameter based on the sixth parameter acquired by the second data acquisition unit.

4. An inference device that uses a learned first model, a learned second model, and a learned third model to infer the maintenance of an air conditioning system including at least one air filter, characterized in that The first model estimates the third parameter based on a first parameter and a second parameter, The second model estimates the fifth parameter based on a fourth parameter, The third model estimates the seventh parameter based on a sixth parameter, The first parameter indicates the degree of clogging of the at least one air filter, The second parameter is a parameter related to the air conditioning intensity of the air conditioning system, The third parameter represents the increase in the power cost of the air conditioning system caused by the first parameter when the second parameter is operating. The fourth parameter represents a first date and time. The fifth parameter represents the air conditioning intensity of the air conditioning system assumed at the first date and time. The sixth parameter represents a second date and time. The seventh parameter represents the maintenance cost of the at least one air filter at the second date and time. The inference device includes: a data acquisition unit that acquires the first parameter, the second parameter, the fourth parameter, and the fifth parameter; and an inference unit that uses the first model, the second model, and the third model. The inference unit uses the first model to estimate the third parameter based on the first parameter and the second parameter, uses the second model to estimate the fifth parameter based on the fourth parameter, and uses the third model to estimate the seventh parameter based on the sixth parameter.

5. The inference device according to claim 4, wherein: the first model, the second model, and the third model are each generated by supervised learning.

Images