WO2021176700A1

WO2021176700A1 - Learning device, inference device, and air conditioner

Info

Publication number: WO2021176700A1
Application number: PCT/JP2020/009746
Authority: WO
Inventors: 哲矢山下
Original assignee: 三菱電機株式会社
Priority date: 2020-03-06
Filing date: 2020-03-06
Publication date: 2021-09-10

Abstract

These learning device and air conditioner each comprise a data acquisition unit for acquiring user response information with respect to an operation and weather information during operation, and a model generation unit for generating a learned model from the weather information and the user response information, using learning data containing the user response information and the weather information. This inference device comprises a data acquisition unit for acquiring current weather information, and an inference unit for inferring an air-conditioning load of an air conditioner from the current weather information acquired by the data acquisition unit, using a learned model generated from learning data containing user response information with respect to an operation and the weather information during operation.

Description

Learning device, inference device and air conditioner

The present disclosure relates to a learning device, an inference device, and an air conditioner, and particularly to an air conditioning load prediction.

The air conditioner controls the rotation speed of the compressor, the rotation speed of the outdoor fan, the rotation speed of the indoor fan, etc. according to the air conditioning load. The air conditioning load is affected by the difference between the outside air temperature and the room temperature, the number of people in the air conditioning space, and the like. In recent years, a method of predicting an air conditioning load by learning using a neural network model (Neural Network Model) has been known (see, for example, Patent Document 1). Patent Document 1 discloses that the number of people in the room and weather information are used as input data for learning.

Japanese Unexamined Patent Publication No. 2006-078009

When a model is used in which only the indoor environmental conditions such as the number of people in the room and the external conditions such as weather information are used as input data as in Patent Document 1, the user's feeling of comfort is sufficient for determining the load conditions. It is not reflected in the above, and it is not possible to improve the comfort of the user depending on the weather conditions.

The present disclosure has been made to solve the above problems, and an object of the present disclosure is to provide a learning device, an inference device, and an air conditioner that improve user comfort regardless of weather conditions.

The learning device according to the present disclosure includes information on the air conditioning load of the air conditioner, user response information obtained from a user command input by the user for the operation of the air conditioner under the air conditioning load, and the air conditioner. Using the data acquisition unit for acquiring the weather information during the operation in the installation environment of the above, and the learning data including the air conditioning load information, the user response information, and the weather information, the weather information and the user. It includes a model generation unit that generates a trained model for inferring the air conditioning load of the air conditioner from the response information.
Further, the inference device according to the present disclosure includes a data acquisition unit that acquires current weather information in the installation environment of the air conditioner, information on the air conditioning load of the air conditioner, and the air conditioner under the air conditioning load. Using a trained model generated by learning from learning data including user response information obtained from a user command input by the user for driving and weather information during the driving in the installation environment of the air conditioner. It also includes an inference unit that infers the air conditioning load of the air conditioner from the current weather information acquired by the data acquisition unit.
Further, the air conditioner according to the present disclosure includes a compressor, an outdoor fan, an indoor fan, the above-mentioned inference device, and the number of rotations of the compressor according to the air conditioning load inferred by the inference device. A control device for controlling at least one of the rotation speed of the outdoor fan and the rotation speed of the indoor fan is provided.

According to the present disclosure, by generating a trained model from learning data including user response information, it is possible to predict an air conditioning load that reflects how the user feels comfort, and the user's comfort regardless of weather conditions. The sex can be improved.

FIG. 5 is a functional block diagram showing a state in which the learning device according to the first embodiment is connected to the Internet and an air conditioner. FIG. 5 is a schematic diagram of a neural network illustrating a method of learning processing performed by the learning device of FIG. 1. It is a flowchart of the learning process performed by the learning apparatus of FIG. FIG. 5 is a functional block diagram showing a state in which the inference device according to the second embodiment is connected to a learning device, an Internet, and an air conditioner. It is a flowchart of the inference processing performed by the inference device of FIG. FIG. 5 is a functional block diagram showing a state in which the machine learning system according to the third embodiment is connected to the Internet and an air conditioner. It is a functional block diagram of the machine learning system which concerns on Embodiment 4. FIG.

Hereinafter, preferred embodiments of the air conditioner of the present disclosure will be described with reference to the drawings. In the following drawings, those having the same reference numerals are the same or equivalent thereof, and this shall be common to the entire text of the specification.

Embodiment 1.
FIG. 1 is a functional block diagram of a state in which the learning device 10 according to the first embodiment is connected to the Internet and an air conditioner 3. The learning device 10 performs learning processing based on information on the operation performed in the past in the air conditioner 3 and the weather information Vwp in the installation environment of the air conditioner 3 when the operation is performed, and performs the learning process. The balancer 3 generates a model for predicting the air conditioning load.

<Structure of air conditioner 3>
The air conditioner 3 includes an outdoor unit 31 installed outdoors, an indoor unit 32 installed in a room, a remote controller 39 operated by a user, an indoor temperature sensor 37 installed in the indoor unit 32, and the like. , Air-conditioning the room. The outdoor unit 31 is equipped with a compressor 34 that compresses the refrigerant, an outdoor fan 35 that blows air, a control device 33 that controls the operation of the air conditioner 3, and the like. The indoor unit 32 is equipped with an indoor fan 36 or the like that blows air.

The outdoor unit 31 and the indoor unit 32 are connected by a signal line or the like so that data can be transmitted and received. Further, the indoor unit 32 and the remote controller 39 transmit and receive data via wireless communication. The user command input to the indoor unit 32 via the remote controller 39 is transmitted from the indoor unit 32 to the control device 33 of the outdoor unit 31. The room temperature sensor 37 detects the room temperature, that is, the room temperature. The detected indoor temperature information is transmitted to the control device 33 of the outdoor unit 31 via the indoor unit 32.

When the air conditioner 3 starts operation and during operation, the control device 33 sets the load condition Va1 such as the rotation speed of the compressor 34, the rotation speed of the indoor fan 36, and the rotation speed of the outdoor fan 35 according to the air conditioning load Va0. And drive. The air conditioning load Va0 is determined, for example, from the difference between the set temperature and the room temperature. Further, when a user command is input via the remote controller 39 during operation, the control device 33 changes the load condition Va1 according to the received user command, and rotates the compressor 34 according to the changed load condition Va1. The number, the rotation speed of the indoor fan 36, and the rotation speed of the outdoor fan 35 are controlled. User commands include, for example, a command for changing the set temperature, a command for changing the strength or direction of the wind blown from the indoor unit 32, and the like.

<Structure of learning device 10>
The learning device 10 is communicably connected to each of the Internet and the air conditioner 3 so that various information can be received from the Internet and the air conditioner 3. The learning device 10 includes a data acquisition unit 11 that acquires information and the like regarding the past operation of the air conditioner 3, a model generation unit 12 that generates a model for predicting an air conditioning load by learning from various acquired information, and a model generation unit 12. It is composed of a model storage unit 13 that stores the generated model.

The learning device 10 is composed of dedicated hardware or a CPU (Central Processing Unit) that executes a program stored in a memory. The CPU is also referred to as a central processing unit, a processing unit, an arithmetic unit, a microprocessor, a microcomputer, or a processor.

When the learning device 10 is dedicated hardware, the learning device 10 may be, for example, a single circuit, a composite circuit, an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or a combination thereof. Applicable. Each of the functional units realized by the learning device 10 may be realized by individual hardware, or each functional unit may be realized by one hardware.

When the learning device 10 is a CPU, each function executed by the learning device 10 is realized by software, firmware, or a combination of software and firmware. Software and firmware are written as programs and stored in memory. The CPU realizes each function of the learning device 10 by reading and executing a program stored in the memory. Here, the memory is a non-volatile or volatile semiconductor memory such as, for example, RAM, ROM, flash memory, EPROM, or EEPROM.

A part of the function of the learning device 10 may be realized by dedicated hardware, and a part may be realized by software or firmware.

The data acquisition unit 11 acquires various information necessary for learning of the model generation unit 12 and creates learning data VL. The data acquisition unit 11 acquires information on the operation performed in the past in the air conditioner 3 from the air conditioner 3, and the weather information in the installation environment of the air conditioner 3 when the operation is performed, that is, the past weather. Obtain information Vwp from the Internet. Here, the operation is defined by the settings used for determining the load conditions such as the set temperature, the strength of the wind blown from the indoor unit 32, and the wind direction. Further, the period during which these settings are constant is the length of the operation. The data acquisition unit 11 creates learning data VL from the acquired information on the past operation of the air conditioner 3 and the weather information Vwp at the time of operation, and outputs the learning data VL to the model generation unit 12.

Here, the information regarding the past operation of the air conditioner 3 includes the air conditioning load Va0 of the operation before the user command is input, the load condition Va1 set in the operation, and the input user command, that is, immediately before. User response information Vres or the like representing the user's response to the operation of the vehicle.

The user response information Vres is obtained from a user command input via the remote controller 39. The user response information Vres is, for example, the time from the start to the stop of the immediately preceding operation, the time from the stop to the start of the operation, the selection result of the operation mode, the time until the operation mode is changed, the time until the set temperature is changed, and the setting. At least one of the temperature change range and the set temperature.

The data acquisition unit 11 creates learning data VL from the acquired air conditioning load Va0 and load condition Va1 information, user response information Vres, and weather information Vwp during operation by normalization or the like. The reason for performing such preprocessing is to convert various information having different scales into data having a common scale and then give it to the model generation unit 12 as input data. As for the method of creating the learning data VL from the various acquired information, a known method may be applied, and the description thereof will be omitted.

As described above, the learning data VL includes elements such as the air conditioning load Va0 during operation, the load condition Va1 during operation, the weather information Vwp during operation, and the user response information Vres for the operation. A plurality of learning data VLs are created each time the operation is changed via the remote controller 39.

Meteorological information Vwp means at least one of temperature, humidity and weather conditions in the installation environment of the air conditioner 3, change status of each condition, or prediction of each condition. Since such weather conditions affect the physical condition and clothes of the user, it is easily imagined that the air conditioning load of the air conditioner 3 is affected by the above weather conditions. Therefore, the learning data VL is configured to include the weather information Vwp as one element.

For example, if the user response information Vres is the time from the start to the stop of the immediately preceding operation, the time from the stop to the start of the operation, the time until the operation mode is changed, or the time until the set temperature is changed, up to the user command. The user's comfort can be judged by the length of time. Specifically, when these times are long, the user feels more comfortable than when they are short, and it can be seen that the air conditioner 3 can be operated under load conditions suitable for the user at this time.

Further, for example, when the user response information Vres is the selection result of the operation mode or the change range of the set temperature, the comfort of the user can be judged by the degree of change from the immediately preceding operation. Specifically, when the user response information Vres is the change range of the set temperature, the user feels more comfortable when the change range is small than when it is large, and the air is air-conditioned under the load conditions suitable for the user at this time. It can be seen that the air conditioner 3 is operating. As described above, the weather conditions affect the physical condition and clothes of the user, and the way the user feels hot and cold differs depending on the weather conditions.

As shown in FIG. 1, the model generation unit 12 learns the air conditioning load based on the learning data VL output from the data acquisition unit 11 and generates a trained model. Further, the model generation unit 12 outputs the generated learned model to the model storage unit 13. In the example shown in FIG. 1, the learning device 10 is configured as a device separate from the air conditioner 3 and is connected to the air conditioner 3 via a network. Here, the learning device 10 may be built on a cloud server. The learning device 10 may be built in the air conditioner 3.

FIG. 2 is a schematic diagram of a neural network illustrating a method of learning processing performed by the learning device 10 of FIG. As the learning algorithm used by the model generation unit 12, known algorithms such as supervised learning, unsupervised learning, and reinforcement learning can be used. Here, supervised learning is to learn the features in the learning data VL by giving a set of data of input and result (also referred to as a label) to the learning device 10. The generated model is used to infer the result from the input. Further, there are various machine learning methods such as learning the air conditioning load of the air conditioner 3, and any method may be used. Hereinafter, a case where the model generation unit 12 learns the air conditioning load by supervised learning according to the neural network model will be described.

The neural network is composed of an input layer composed of a plurality of neurons X1 to X3, an intermediate layer composed of a plurality of neurons Y1 to Y2 (also referred to as a hidden layer), and an output layer composed of a plurality of neurons Z1 to Z3. The intermediate layer may be one layer or two or more layers. The weights w11 to w16 indicate the strength of the connection between the neurons X1 to X3 in the input layer and the neurons Y1 to Y2 in the intermediate layer. The weights w21 to w26 indicate the strength of the connection between the neurons Y1 to Y2 in the intermediate layer and the neurons Z1 to Z3 in the output layer.

For example, in a neural network having one intermediate layer and a total of three layers as shown in FIG. 2, when a plurality of inputs are input to the input layer, the values are multiplied by the weights w11 to w16 and input to the intermediate layer. , The result of the intermediate layer is further multiplied by the weights w21 to w26 to be output from the output layer. This output result changes depending on the values of the weights w11 to 16 and w21 to 26.

In the present disclosure, the neural network learns the air conditioning load by so-called supervised learning according to the learning data VL created based on the combination of the weather information Vwp and the user response information Vres acquired by the data acquisition unit 11. .. Specifically, the neural network is weighted so that the result output from the output layer by inputting the weather information Vwp to the input layer approaches the air conditioning load to which the user response information Vres is applied to the operation immediately before the user command. Adjust the value.

As described above, since the user response information Vres for driving is incorporated in the learning data VL together with the weather information Vwp during operation in the installation environment of the air conditioner 3, the trained model is based on the weather information Vww. It reflects how the user feels.

As shown in FIG. 1, the trained model generated as described above is stored in the model storage unit 13 as a trained model for the air conditioner 3. The trained model stored in the model storage unit 13 can be referred to and updated by the model generation unit 12 as needed.

FIG. 3 is a flowchart of the learning process performed by the learning device 10 of FIG. The learning process performed by the learning device 10 will be described with reference to FIGS. 1 and 3. The data acquisition unit 11 acquires the user response information Vres, the air conditioning load Va0 and the load condition Va1 of the immediately preceding operation from the air conditioner 3, and the weather information Vwp during operation in the installation environment of the air conditioner 3 from the Internet. (Step ST101). It is assumed that the weather information Vwp and the user response information Vres in the installation environment of the air conditioner 3 are acquired at the same time, but if the weather information Vwp and the user response information Vres are input in association with each other, the timings are different. May be obtained at. The data acquisition unit 11 creates learning data VL from the acquired information (step ST102) and outputs the learning data VL to the model generation unit 12. The model generation unit 12 learns the air conditioning load from the input learning data VL and generates a trained model (step ST103). The model generation unit 12 outputs the generated trained model to the model storage unit 13, and the model storage unit 13 stores the trained model (step ST104).

The model generation unit 12 may learn the air conditioning load according to the learning data VL created for the plurality of air conditioners 3. The model generation unit 12 may acquire learning data from a plurality of air conditioners 3 used in the same area, or may acquire learning data from a plurality of air conditioners operating independently in different areas. May be acquired and used for learning. It is also possible to add or remove the air conditioner for which the learning data is to be collected from the target during the operation.

As described above, the learning device 10 of the first embodiment includes a data acquisition unit 11 and a model generation unit 12. The data acquisition unit 11 includes information on the air conditioning load Va0 of the air conditioner 3, user response information Vres obtained from a user command input by the user for the operation of the air conditioner 3 under the air conditioning load, and the air conditioner 3. Acquires the weather information Vwp during operation in the installation environment of. The model generation unit 12 infers the air conditioning load of the air conditioner 3 from the weather information Vwp and the user response information Vres by using the learning data VL including the information of the air conditioning load Va0, the user response information Vres, and the weather information Vwp. Generate a trained model to do.

As a result, the trained model is generated from the learning data VL including the user response information Vres, so that it is possible to predict the air conditioning load that reflects the user's feeling of comfort, and the user's comfort regardless of the weather conditions. The sex can be improved.

Further, the air conditioner 3 of the first embodiment may include a compressor 34, an outdoor fan 35, an indoor fan 36, the above-mentioned inference device 20, and a control device 33. The control device 33 controls at least one of the rotation speed of the compressor 34, the rotation speed of the outdoor fan 35, and the rotation speed of the indoor fan 36 according to the air conditioning load inferred by the inference device 20.

As a result, the air conditioner 3 can be operated with an air-conditioned load that reflects the user's feeling of comfort, and the user's comfort can be improved regardless of the weather conditions. Further, by performing such an operation, the number of operations by the user can be reduced as compared with the case where the operation is performed with the air conditioning load predicted by the conventional method.

Further, the air conditioner 3 may further include the above-mentioned learning device 10. As a result, the learning device 10 can acquire information on the operation of the air conditioner 3 in the air conditioner 3, so that the learned model can be easily updated.

Embodiment 2.
FIG. 4 is a functional block diagram of a state in which the inference device 20 according to the second embodiment is connected to the learning device 10, the Internet, and the air conditioner 3. The inference device 20 of the second embodiment infers the optimum air conditioning load of the air conditioner 3 with respect to the weather information Vwn by using the learned model generated by the learning device 10 of the first embodiment. .. As shown in FIG. 4, the inference device 20 of the second embodiment and the learning device 10 of the first embodiment predict the air conditioning load of the present or the next day, and the optimum load for the air conditioner 3. A machine learning system 1 that provides the condition Va21 is formed.

<Configuration of inference device 20>
The inference device 20 is communicably connected to each of the Internet and the air conditioner 3 so that various information can be received from the Internet and the air conditioner 3. Further, the inference device 20 is communicably connected to the learning device 10. The inference device 20 infers the air conditioning load from the acquired information by using the data acquisition unit 21 that acquires information about the current operation from the air conditioner 3 and the learned model generated by the learning device 10. It is composed of a part 22 and a part 22. The case of optimizing the current air conditioning load will be described below.

The inference device 20 is composed of dedicated hardware or a CPU (Central Processing Unit) that executes a program stored in a memory. The CPU is also referred to as a central processing unit, a processing unit, an arithmetic unit, a microprocessor, a microcomputer, or a processor.

When the inference device 20 is dedicated hardware, the inference device 20 may be, for example, a single circuit, a composite circuit, an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or a combination thereof. Applicable. Each of the functional units realized by the inference device 20 may be realized by individual hardware, or each functional unit may be realized by one hardware.

When the inference device 20 is a CPU, each function executed by the inference device 20 is realized by software, firmware, or a combination of software and firmware. Software and firmware are written as programs and stored in memory. The CPU realizes each function of the inference device 20 by reading and executing a program stored in the memory. Here, the memory is a non-volatile or volatile semiconductor memory such as, for example, RAM, ROM, flash memory, EPROM, or EEPROM.

A part of the function of the inference device 20 may be realized by dedicated hardware, and a part may be realized by software or firmware.

The data acquisition unit 21 acquires various information necessary for the inference of the inference unit 22. Specifically, the data acquisition unit 21 acquires information on the current air conditioning load Va10 and load condition Va11 of the air conditioner 3 from the air conditioner 3, and obtains the current weather information Vwn in the installation environment of the air conditioner 3. Get from the internet. The data acquisition unit 21 outputs various acquired information to the inference unit 22.

Here, the current air conditioning load Va10 is, for example, the air conditioning capacity required to bring the current indoor temperature close to the set temperature. In the air conditioner 3, the control device 33 determines the load condition Va11 such as the rotation speed of the compressor 34, the rotation speed of the outdoor fan 35, and the rotation speed of the indoor fan 36 according to the current air conditioning load Va10.

The inference unit 22 infers the optimum air conditioning load of the air conditioner 3 for the weather information Vwn by using the learned model generated in advance by the learning device 10 for the air conditioner 3. Specifically, various information including the current weather information Vwn acquired by the data acquisition unit 21 is input to the trained model, and the optimum air conditioning load is output for the air conditioner 3. Further, the inference unit 22 transmits the optimum load condition Va21 according to the air conditioning load obtained by the inference to the air conditioner 3.

In the example shown in FIG. 4, the inference unit 22 infers the air conditioning load using the learned model learned by the learning device 10 of the machine learning system 1 outside the air conditioner 3, and responds to the inferred air conditioning load. The case where the optimum load condition Va21 is output to the air conditioner 3 is shown. Here, the machine learning system 1 may be built on a cloud server. Further, for example, the air conditioner 3 may have a configuration in which a trained model is acquired from another external air conditioner and inference is performed based on the trained model. Even if the learning device 10 that has learned the air conditioning load for a certain air conditioner is applied to an air conditioner different from the air conditioner, and the air conditioning load is relearned and updated in the applied air conditioner. good. The machine learning system 1 may be built in the air conditioner 3.

FIG. 5 is a flowchart of inference processing performed by the inference device 20 of FIG. The learning process performed by the inference device 20 will be described with reference to FIGS. 4 and 5. The data acquisition unit 21 acquires the current air conditioning load Va10 and load condition Va11 from the air conditioner 3, and acquires the current weather information Vwn in the installation environment of the air conditioner 3 from the Internet (step ST201). The timing of acquiring the current weather information Vwn and the current air conditioning load Va10 or the like does not have to be exactly the same as long as the weather information Vwn and the air conditioning load Va10 can be input in association with each other. The data acquisition unit 21 outputs these acquired information to the inference unit 22. The inference unit 22 inputs various information including the current weather information Vwn input from the data acquisition unit 21 into the latest learned model of the air conditioner 3 acquired from the learning device 10 in advance, and air-conditions. Infer the load (step ST202). The inference unit 22 outputs the optimum load condition Va21 according to the air conditioning load obtained by the inference to the air conditioner 3 (step ST203). As the optimum load condition Va21, for example, the rotation speed of the compressor 34, the rotation speed of the indoor fan 36, the rotation speed of the outdoor fan 35, and the like are output in consideration of the user's feeling.

<Operation of air conditioner 3>
Hereinafter, the operation of the air conditioner 3 when the air conditioning load is optimized by the machine learning system 1 when the air conditioner 3 is operated by the automatic operation control will be described. When automatic operation control is selected, the air conditioner 3 determines the required air conditioning load, for example, giving priority to reducing the temperature difference between the room temperature and the set temperature. Specifically, the control device 33 determines the rotation speed of the compressor 34 and the rotation speed of the outdoor fan 35, and the strength of the wind blown from the air conditioner 3, that is, the rotation speed of the indoor fan 36 and the air flow. The orientation of the air conditioner is automatically determined.

The air conditioner 3 outputs a start signal to the inference device 20 when the operation is started by the automatic operation control. First, the air conditioner 3 transmits the air conditioning load Va10 and the load condition Va11 determined by the control device 33 to the inference device 20 in response to the request of the inference device 20 that has received the start signal. After that, the air conditioner 3 receives the optimum load condition Va21 from the inference device 20, and controls the compressor 34, the outdoor fan 35, and the indoor fan 36 according to the received load condition Va21. During the operation under the automatic operation control, the air conditioner 3 repeats the process of transmitting information to the inference device 20 every time the control device 33 determines the air conditioning load and receiving the optimum load condition Va21 from the inference device 20. When the user operates the remote controller 39 during the operation in the automatic operation control, the control device 33 outputs an end signal to the inference device 20, and shifts from the automatic operation control to the user operation control. In the user operation control, for example, the set temperature, the wind direction of the blown airflow, and the wind strength can be set via the remote controller 39.

As described above, the inference device 20 of the second embodiment uses the data acquisition unit 21 for acquiring the current weather information Vwn in the installation environment of the air conditioner 3 and the trained model to perform air conditioning from the weather information Vwn. It is provided with an inference unit 22 for inferring the air conditioning load of the machine 3. In the second embodiment, the trained model used by the inference unit 22 is learning data including user response information Vres for the operation of the air conditioner 3 and weather information Vwp during operation in the installation environment of the air conditioner 3. It was generated by learning from. Therefore, the inference device 20 of the second embodiment has the same effect as the learning device 10 of the first embodiment.

In the example shown in FIG. 4, the learning device 10 and the inference device 20 have been described as separate devices, but the learning device 10 and the inference device 20 may be configured by one controller.

Embodiment 3.
FIG. 6 is a functional block diagram of the machine learning system 1 according to the third embodiment connected to the Internet and the air conditioner 3. In the third embodiment, the machine learning system 1 learns and predicts the air conditioning load in consideration of the indoor environmental condition Volume with respect to the air conditioner 3 having the motion sensor 38a that detects the indoor environmental condition Volume of the air conditioning target space. Is possible.

Here, the indoor environmental condition Volume means at least one of the number of persons in the air-conditioned space, the space temperature, the space humidity, the surface temperature including the people in the air-conditioned space, and the amount of change thereof. The motion sensor 38a is composed of, for example, an infrared sensor or the like. An area sensor may be adopted instead of the motion sensor 38a, or the motion sensor 38a and the area sensor may be used in combination.

<Structure of learning device 10>
The data acquisition unit 11 of the learning device 10 acquires information on the past operation of the air conditioner 3 from the air conditioner 3, and obtains the weather information when the operation is being performed, that is, the past weather information Vwp from the Internet. get. Here, the operation is defined by the settings used for determining the load conditions such as the set temperature, the strength of the wind blown from the indoor unit 32, and the wind direction. Further, the period during which these settings are constant is the length of the operation. The data acquisition unit 11 creates learning data VL from the acquired information on the past operation of the air conditioner 3 and the weather information Vwp at the time of operation, and outputs the learning data VL to the model generation unit 12.

Here, the information on the past operation of the air conditioner 3 also includes the indoor environmental condition Volume obtained by the motion sensor 38a during operation. The data acquisition unit 11 creates learning data VL from the acquired air conditioning load Va0 information, load condition Va1 information, user response information Vres, indoor environment condition Room, and weather information Vwp during operation by normalization or the like. A plurality of learning data VLs are created each time the operation is changed via the remote controller 39.

The model generation unit 12 learns the optimum air conditioning load based on the learning data VL input from the data acquisition unit 11, and generates a trained model. Further, the model generation unit 12 outputs the generated learned model to the model storage unit 13.

Learning is performed by using the learning data VL as so-called teacher data, for example, by using a neural network. Specifically, the neural network inputs the weather information Vwp and the indoor environmental condition Room to the input layer, and the result output from the output layer is applied to the air conditioning load to which the user response information Vres is applied to the operation immediately before the user command. Adjust the weight values so that they are closer (see FIG. 2). As described above, in the third embodiment, the learning data VL includes the weather information Vwp during operation in the installation environment of the air conditioner 3, the user response information Vres for operation, and the indoor environmental condition Volume that affects the air conditioning load. Is built in. Therefore, the trained model reflects the user's feelings about the weather condition and the indoor environmental condition Bloom.

<Configuration of inference device 20>
The data acquisition unit 21 of the inference device 20 acquires information on the current operation of the air conditioner 3 from the air conditioner 3, and acquires the current weather information Vwn in the installation environment of the air conditioner 3 from the Internet. Here, the information regarding the current operation of the air conditioner 3 includes the current indoor environmental condition Room. The data acquisition unit 21 outputs the acquired information on the current air conditioning load Va0 and the load condition Va1, the current indoor environmental condition Room, and the weather information Vwn to the inference unit 22.

The inference unit 22 infers the optimum air conditioning load of the air conditioner 3 for the weather condition and the indoor environmental condition Room by using the learned model generated in advance by the learning device 10 for the air conditioner 3. Specifically, information including the current weather information Vwn and the current indoor environmental condition Bloom acquired by the data acquisition unit 21 is input to the trained model, and the optimum air conditioning load for the air conditioner 3 is inferred. The inference unit 22 transmits the load condition Va21 according to the air conditioning load obtained by the inference to the air conditioner 3.

As described above, the air conditioner 3 includes a sensor 38a that detects the indoor environmental condition Volume of the air-conditioned space, and the learning data VL is the indoor environmental condition Volume detected by the motion sensor 38a of the air conditioner 3. Is included. Further, the indoor environmental condition Volume includes at least one of the number of persons in the air-conditioned space, the space temperature, the space humidity, and the surface temperature including the person in the air-conditioned space.

As a result, the trained model can be enhanced by combining the weather information Vwn and Vwp in the installation environment of the air conditioner 3 with the indoor environment condition Room in which the user is present, and more accurate air conditioning load prediction according to the indoor environment. Is possible.

Embodiment 4.
FIG. 7 is a functional block diagram of the machine learning system 1 according to the fourth embodiment. In the fourth embodiment, the machine learning system 1 can learn and predict the air conditioning load according to the detected user with respect to the air conditioner 3 provided with the user detection device that detects the user in the room. There is.

Here, the user detection device is composed of the indoor unit 32 or the user detection unit 38b installed in the room, and the control device 33. The user detection unit 38b includes, for example, at least one of a motion sensor, a voice recognition device, a mobile terminal, and a wearable terminal, and the detection information is transmitted to the control device 33. With such a configuration, the user can be identified by a method such as image analysis of the indoor space, voice recognition by voice, or analysis of information transmitted from the terminal. Then, the user does not need to carry a card or the like for identification, and the air conditioner suitable for himself / herself can be obtained only by carrying it with what he / she usually has.

User information Vu corresponding to a plurality of users is registered in advance in the control device 33 of the air conditioner 3, and user information Vu can be added to and deleted from the control device 33. User information Vu is information that identifies a user. The control device 33 has a function of identifying one or more users in the room based on the detection information of the user detection unit 38b, and outputs the user information Vu corresponding to the users in the room. Specifically, when the air conditioner 3 is operated, the control device 33 stores the user information Vu of the user in the room as one of the information related to the operation, and transmits the user information Vu in response to an external request. The user detection device may be configured by the user detection unit 38b alone, as long as it can identify one or two or more users in the room.

Here, the information regarding the past operation of the air conditioner 3 includes information on the air conditioning load Va0 during operation, information on the load condition Va1 during operation, user response information Vres, and is detected by the user detection device during operation. The information of the user, that is, the user information Vu is included. A plurality of user information Vu can be registered in the machine learning system 1. The data acquisition unit 11 creates learning data VL from the acquired air conditioning load Va0 information, load condition Va1 information, weather information Vwp, and user response information Vres by normalization or the like. A plurality of learning data VLs are created each time the operation is changed via the remote controller 39. The data acquisition unit 11 divides a plurality of learning data VL for each user based on the acquired user information Vu and outputs the plurality of learning data VL to the model generation unit 12.

The model generation unit 12 learns the optimum air conditioning load for the learning data VLa, VLb, and VLc of each user input from the data acquisition unit 11, and generates the trained models A, B, and C for each user. Further, the model generation unit 12 outputs each generated model to the model storage unit 13.

As shown in FIG. 7, in the model storage unit 13, the trained models A, B, and C generated for the three users as described above are stored as trained models for the air conditioner 3. .. Each trained model stored in the model storage unit 13 can be referred to and updated by the model generation unit 12 as needed.

<Configuration of inference device 20>
The data acquisition unit 21 of the inference device 20 acquires information on the current operation of the air conditioner 3 from the air conditioner 3, and acquires the current weather information Vwn in the installation environment of the air conditioner 3 from the Internet. Here, the information regarding the operation currently being performed in the air conditioner 3 includes information on the current air conditioning load Va10, information on the current load condition Va1, and information on the user in the room. The data acquisition unit 21 outputs the acquired air conditioning load Va10 information, load condition Va11 information, user information Vu, and weather information Vwn to the inference unit 22.

The inference unit 22 infers the weather conditions and the optimum air conditioning load of the air conditioner 3 for the user by using the learned model generated in advance by the learning device 10 for each user for the air conditioner 3. Specifically, information including the current weather information Vwn is input to the learned model corresponding to the user information Vu input from the data acquisition unit 21, and the air conditioner 3 is used for the user in the room. The optimum air conditioning load is output. The inference unit 22 transmits the load condition Va21 according to the air conditioning load obtained by the inference to the air conditioner 3.

As described above, in the learning device 10 of the fourth embodiment, the data acquisition unit 11 acquires the user information Vu, and the model generation unit 12 generates the learned model according to the user information Vu. Further, in the inference device 20 of the fourth embodiment, the data acquisition unit 21 acquires the user information Vu, and the inference unit 22 uses the learned model corresponding to the user information Vu acquired by the data acquisition unit 21. Infer the air conditioning load of the air conditioner 3.

As a result, the air conditioning load can be predicted according to the user of the air conditioner 3, and air conditioning suitable for each person can be performed among users who feel cold or hot and who have different clothes.

Note that the third embodiment and the fourth embodiment may be combined. Specifically, the model generation unit 12 of the learning device 10 generates a trained model for each user according to the user information Vu from the learning data VL including the indoor environmental condition Bloom detected by the air conditioner 3. You may. Further, the inference unit 22 of the inference device 20 infers the air conditioning load according to the user information Vu by using the learned model generated from the learning data VL including the indoor environmental condition Bloom detected in the air conditioner 3. do. As a result, the operation that suits the user's feeling is further adjusted according to the indoor environmental conditions such as the number of people in the room, and the current air conditioning load that is more suitable for the user is transferred to the air conditioner 3 by the machine learning system 1. It will be possible to provide. Further, in the configuration in which the third embodiment and the fourth embodiment are combined, both the indoor environmental condition Room and the indoor user information Vu may be detected by the same sensor.

It is possible to combine each embodiment and appropriately modify or omit each embodiment. The embodiments of the present disclosure are not limited to the above embodiments, and various modifications can be made. For example, in the above embodiment, the learning device 10 and the inference device 20 are configured to acquire the weather information Vwp and Vwn via the Internet, but the configuration is not particularly limited to such a configuration. For example, when the air conditioner 3 stores the weather information Vwp during operation and the current weather information Vwn, the learning device 10 acquires the weather information Vww during operation from the air conditioner 3 and is an inference device. 20 may acquire the current weather information Vwn from the air conditioner 3. When the outside air temperature is used as the weather information Vwp and Vwn, the outdoor unit 31 of the air conditioner 3 is provided with an outside air temperature sensor or the like for detecting the outside air temperature. The outside air temperature detected by the sensor may be stored.

Further, the learning device 10 may be configured to determine whether or not to use the indoor environmental condition Room for learning based on the device information of the air conditioner 3. Further, the learning method is not limited to the above neural network model, and a known method can be used instead.

1 machine learning system, 3 air conditioner, 10 learning device, 11 data acquisition unit, 12 model generation unit, 13 model storage unit, 20 inference device, 21 data acquisition unit, 22 inference unit, 31 outdoor unit, 32 indoor unit, 33 control device, 34 compressor, 35 outdoor fan, 36 indoor fan, 37 indoor temperature sensor, 38a human sensor, 38b user detector, 39 remote controller, Va1, Va11, Va21 load condition, VL, VLa, VLb, VLc learning Data, Va0, Va10 air conditioning load, Vres user response information, Room indoor environmental conditions, Vu user information, Vwn, Vwp weather information.

Claims

Information on the air conditioning load of the air conditioner, user response information obtained from a user command input by the user for the operation of the air conditioner under the air conditioner load, and information on the operation of the air conditioner in the installation environment of the air conditioner. Meteorological information, data acquisition department to acquire, and
Using the learning data including the air conditioning load information, the user response information, and the weather information, a trained model for inferring the air conditioning load of the air conditioner from the weather information and the user response information is obtained. The model generator to generate and
A learning device equipped with.
The data acquisition unit that acquires the current weather information in the installation environment of the air conditioner,
Information on the air conditioning load of the air conditioner, user response information obtained from a user command input by the user for the operation of the air conditioner under the air conditioning load, and the operation of the air conditioner in the installation environment of the air conditioner. The inference unit that infers the air conditioning load of the air conditioner from the current weather information acquired by the data acquisition unit using the trained model generated by learning from the training data including the weather information of the above.
Inference device equipped with.
The learning device according to claim 1, wherein the weather information includes at least one of temperature, humidity, and weather conditions in the installation environment of the air conditioner, a change state of each of the conditions, and a prediction of each of the conditions. , Or the inference device according to claim 2.
The user response information includes the time from the start to the stop of the operation of the air conditioner, the time from the stop to the start of the operation, the selection result of the operation mode, the time until the change of the operation mode, and the set temperature. The learning device according to claim 1, or the inference device according to claim 2, which includes at least one of the time until the change, the change width of the set temperature, and the set temperature.
The air conditioner includes a sensor that detects the indoor environmental conditions of the air-conditioned space.
The learning device according to claim 1, or the inference device according to claim 2, wherein the learning data includes the indoor environmental conditions detected by the sensor of the air conditioner.
The learning device according to claim 5, wherein the indoor environmental condition includes at least one of the number of persons, the space temperature, the space humidity, and the surface temperature including the person in the air-conditioned space. The inference device according to 5.
The data acquisition unit acquires user information, which is information that identifies the user, and obtains user information.
The learning device according to claim 1, wherein the model generation unit generates the trained model according to the user information.
The data acquisition unit acquires user information, which is information that identifies the user, and obtains user information.
The inference device according to claim 2, wherein the inference unit infers the air conditioning load of the air conditioner by using the learned model corresponding to the user information acquired by the data acquisition unit.
The air conditioner includes a user detection device that detects a user in the air-conditioned space.
The learning device according to claim 7, wherein the user detection device has at least one of a motion sensor, a voice recognition device, a mobile terminal, and a wearable terminal, and outputs the user information corresponding to the detected user. Alternatively, the inference device according to claim 8.
The air conditioner includes a sensor that detects the indoor environmental conditions of the air-conditioned space.
The learning data includes the indoor environmental conditions detected by the sensor of the air conditioner.
The data acquisition unit acquires user information and
The learning device according to claim 1, wherein the model generation unit generates the trained model according to the user information.
The air conditioner includes a sensor that detects the indoor environmental conditions of the air-conditioned space.
The learning data includes the indoor environmental conditions detected by the sensor of the air conditioner.
The data acquisition unit acquires user information and
The inference device according to claim 2, wherein the inference unit infers the air conditioning load of the air conditioner by using the learned model corresponding to the user information acquired by the data acquisition unit.
With a compressor,
With an outdoor fan
With an indoor fan
The inference device according to claim 2 and
A control device that controls at least one of the rotation speed of the compressor, the rotation speed of the outdoor fan, and the rotation speed of the indoor fan according to the air conditioning load inferred by the inference device.
Air conditioner equipped with.
The air conditioner according to claim 12, further comprising the learning device according to claim 1.