Air conditioner

Abstract

To provide an air conditioner which enables a user of the air conditioner to receive the benefits by a learning model in an early stage.SOLUTION: An air conditioner comprises an adapter which communicates with a server device. The adapter includes a storage section, a communication section, and a control section. The storage section stores an initial learning model having a learned parameter relating to predetermined prediction processing to be executed by the air conditioner. The communication section collects sensor data detected by a sensor that the air conditioner has, transmits the sensor data to the server device, and receives a first learning model which is generated by the server device on the basis of the sensor data. The control section executes the predetermined prediction processing using the initial learning model until receiving the first learning model, and executes the predetermined prediction processing using the first learning model after the first learning model is received.SELECTED DRAWING: Figure 5

Description

[Technical field]

[0001] The present invention relates to an air conditioner capable of performing various prediction functions through learning processing. [Background technology]

[0002] Conventionally, there has been known an air conditioning system that generates a learning model by having a learning unit such as an AI (Artificial Intelligence) learn operation history data such as temperature settings, and controls air conditioning operation using the generated learning model (hereinafter also referred to as AI control). The learning model generated by the learning unit has parameters obtained by learning (also referred to as learned parameters).

[0003] For example, the following Patent Document 1 describes an air conditioning system equipped with a cloud-side learning unit and an adapter-side learning unit. Specifically, the adapter downloads a first learning model generated by the cloud-side learning unit and sets it as a second learning model, and the adapter-side learning unit performs learning on the second learning model based on the most recent operation history data (user's operating preferences in the air-conditioned space and changes in the actual environment) to generate a new second learning model. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent Publication No. 2021-63611 Summary of the Invention [Problem to be solved by the invention]

[0005] However, even in the case of an air conditioner that combines rapid learning on the air conditioner side with highly accurate learning on the cloud side as described in Patent Document 1, highly accurate learning requires the collection of data indicating user preferences and the presence or absence of users in the air-conditioned space over a certain period of time after the air conditioner is installed. For this reason, users of the air conditioner cannot enjoy the benefits of the learning model early on until the data is collected.

[0006] In view of the above circumstances, an object of the present invention is to provide an air conditioner that enables users of the air conditioner to quickly receive the benefits of a learning model. [Means for solving the problem]

[0007] An air conditioner according to one embodiment of the present invention is an air conditioner equipped with an adapter that communicates with a server device. The adapter has a storage unit, a communication unit, and a control unit. The storage unit stores an initial learning model having learned parameters related to a predetermined prediction process executed by the air conditioner. The communication unit collects sensor data detected by a sensor included in the air conditioner and transmits it to the server device, and receives a first learning model generated by the server device based on the sensor data. The control unit executes the predetermined prediction process using the initial learning model until it receives the first learning model, and executes the predetermined prediction process using the first learning model after it receives the first learning model.

[0008] With this configuration, prediction processing is performed using the initial learning model until the first learning model is received, allowing users of the air conditioner to benefit from the learning model at an early stage.

[0009] The control unit may update the learned parameters of the initial learning model based on the sensor data until the first learning model is received, and after the first learning model is received, update the learned parameters of the first learning model based on the sensor data.

[0010] With this configuration, the learning model can be additionally trained using sensor data on the air conditioner side, regardless of whether the first learning model has been received or not.

[0011] The control unit may predict the presence or absence of a user of the air conditioner in the air-conditioned space using the initial learning model and the first learning model.

[0012] With this configuration, the presence or absence of a user can be predicted even before the first learning model is received.

[0013] The initial learning model may have a plurality of the learned parameters to accommodate a plurality of presence / absence tendencies of the occupant in the air-conditioned space.

[0014] With this configuration, the user's presence or absence can be predicted based on multiple presence / absence patterns even before receiving the first learning model.

[0015] The multiple learned parameters may include a first learned parameter corresponding to a presence / absence pattern having a first absence time as an absence time in the air-conditioned space, and a second learned parameter corresponding to a presence / absence pattern having a second absence time shorter than the first absence time as an absence time in the air-conditioned space.

[0016] With this configuration, even before receiving the first learning model, the presence / absence of a user can be predicted according to a presence / absence pattern in which the user is out for a long time and a presence / absence pattern in which the user is out for a short time, for example, on weekdays and holidays.

[0017] The control unit may start collecting the sensor data when external power is supplied to the air conditioner, and may continue collecting the sensor data regardless of whether air conditioning operation of the air conditioner is started or stopped.

[0018] With this configuration, as long as external power is being supplied to the air conditioner, sensor data can be collected even when the air conditioner is not performing air conditioning operation.

[0019] The control unit may compare the received first learning model with the previously received first learning model each time it receives the latest first learning model from the server device, and if the two models do not match, update the previously received first learning model and a second learning model obtained by additionally learning the previously received first learning model based on the sensor data using the latest first learning model, store them in the memory unit, and execute the specified prediction process using the updated second learning model.

[0020] With this configuration, the previously received first learning model is stored for comparison, and it is determined whether there is a difference between the latest first learning model and the previously received first learning model, making it possible to always use the latest second learning model while avoiding unnecessary update processing.

[0021] The sensor data may be set to be collected at a predetermined time interval, have a timestamp indicating the collection time, and be accumulated in the order of the timestamp. In this case, the control unit may determine whether a timestamp of a predetermined sensor data from among the sensor data having the predetermined timestamp is within a predetermined time range before and after a timestamp that the sensor data acquired in the predetermined order should have when the setting is followed, and may determine that the sensor data is missing when the control unit determines that the timestamp is outside the predetermined time range before and after.

[0022] With this configuration, even if there is an error in the timestamp of the sensor data, it is possible to reliably detect defects (missing parts) of the sensor data. As a result, in this embodiment, it is possible to stop additional learning using a dataset in which sensor data is considered to be missing. If there is no missing part in the dataset collected thereafter (for example, the dataset from the next day onwards), additional learning is performed using the subsequent dataset. Effect of the Invention

[0023] As described above, according to the present invention, it is possible to provide an air conditioner that allows a user of the air conditioner to quickly receive the benefits of a learning model. However, this effect does not limit the present invention. [Brief description of the drawings]

[0024] [Figure 1] 1 is a diagram showing the configuration of an air conditioning system according to an embodiment of the present invention. [Diagram 2] 1 is a block diagram showing a configuration of an adapter for an air conditioner according to an embodiment of the present invention. [Diagram 3] 1 is a block diagram showing a configuration of a server device of an air conditioning system according to an embodiment of the present invention. [Figure 4] 10 is a flowchart showing a flow of a learning process by a server device in one embodiment of the present invention. [Diagram 5] 11 is a flowchart showing the flow of presence / absence prediction processing by an adapter of an air conditioner in one embodiment of the present invention. [Figure 6] 11 is a diagram illustrating pre-processing of sensor data by an adapter of an air conditioner according to an embodiment of the present invention. FIG. [Figure 7] A figure showing an update formula for the presence / absence pattern of the learning model by the adapter of the air conditioner in one embodiment of the present invention. [Figure 8] 11 is a flowchart showing the flow of a learning model update process by an adapter of an air conditioner in one embodiment of the present invention. [Figure 9] A figure explaining the update process of a learning model by an adapter of an air conditioner in one embodiment of the present invention. [Figure 10] A diagram showing a schematic diagram of the presence / absence patterns of the learning model used by the air conditioner of one embodiment of the present invention. [Figure 11] A diagram showing a schematic diagram of an initial presence / absence pattern of an initial learning model used by an air conditioner according to one embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0025] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0026] [System configuration] FIG. 1 is a diagram showing the configuration of an air conditioning system according to this embodiment.

[0027] As shown in the figure, the air conditioning system according to this embodiment has an air conditioner 100, an access point 3, a server device 4, a relay device 5, and a communication terminal 6, which are connected via a network 7. The air conditioner 100 has an indoor unit 1 and an adapter 2.

[0028] The indoor unit 1 is, for example, a part of an air conditioner 100 that is placed indoors and heats or cools the air in the room. A user of the indoor unit 1 can remotely control the indoor unit 1 by operating a remote control 8. The indoor unit 1 is connected to an outdoor unit (not shown) by refrigerant piping, and the outdoor unit is equipped with an outdoor fan, a compressor, etc.

[0029] The indoor unit 1 has a main body 1A, a control unit 1B that controls the main body 1A, and a sensor 1C. The main body 1A has various devices such as an indoor fan, an expansion valve, an indoor heat exchanger, and a wind direction plate (not shown) housed in a housing.

[0030] The control unit 1B controls the indoor fan, expansion valve, and air deflector. The control unit 1B adjusts the opening of the expansion valve to allow the amount of refrigerant required to provide the air conditioning capacity required by the user to flow into the indoor heat exchanger. The control unit 1B also drives the indoor fan to control the air deflector to deflect the indoor air that has exchanged heat with the refrigerant in the indoor heat exchanger and blow it out into the air-conditioned space. This allows the air-conditioned space to be heated, cooled, and dehumidified.

[0031] For example, when the indoor unit 1 is installed and then connected to a commercial power source to receive power, the sensor 1C starts detecting the presence or absence of people in the air-conditioned space in which the indoor unit 1 is installed. Note that from this point on, unless the power supply is interrupted, the sensor 1C continues to detect the presence or absence of people in the air-conditioned space regardless of whether the indoor unit 1 is operating or not.

[0032] Other examples of the sensor 1C that may be provided include a radiation sensor that measures the radiation temperature of the floor surface of the air-conditioned space, a room temperature sensor that measures the indoor temperature of the air-conditioned space, and a humidity sensor that measures the indoor humidity of the air-conditioned space.

[0033] The adapter 2 has a communication function for connecting the indoor unit 1 and the access point 3 via wireless communication, and a control function for AI-controlling the indoor unit 1. The adapter 2 is provided for each indoor unit 1.

[0034] The access point 3 is a device that connects the adapter 2 to a network 7 by wireless communication using, for example, a wireless local area network (WLAN) etc. The network 7 is, for example, a communication network such as the Internet.

[0035] The server device 4 has a function of generating a learning model used when the adapter 2 performs AI control on the indoor unit 1, a database for storing sensor data, etc. The server device 4 is disposed in, for example, a data center.

[0036] The relay device 5 is connected to the network 7 via wired communication, and is also connected to the server device 4 via wired communication. The relay device 5 receives sensor data and the like used to generate a learning model or update a learning model in the server device 4 from the indoor unit 1 via the adapter 2 and the network 7, and transmits the data to the server device 4. The relay device 5 also transmits a learning model generated or updated in the server device 4 to the adapter 2 via the network 7. The relay device 5 is disposed, for example, in a data center or the like. The relay device 5 has a first relay unit 5A, a second relay unit 5B, and a third relay unit 5C.

[0037] The first relay unit 5A relays various data related to AI control between the adapter 2 and the server device 4. Specifically, the first relay unit 5A transmits sensor data and the like used for generating or updating a learning model received from the adapter 2 via the access point 3 and the network 7 to the server device 4, and transmits the learning model generated or updated by the server device 4 to the adapter 2 via the network 7 and the access point 3.

[0038] The second relay unit 5B receives the operating conditions of the indoor unit 1 (such as the operating mode such as cooling / heating and the set temperature) that the user has set using the communication terminal 6 while away from home via the access point 3 and the network 7, and transmits these to the indoor unit 1 via the network 7, the access point 3, and the adapter 2.

[0039] The third relay unit 5C receives external data such as a weather forecast, for example, via the network 7, and transmits the received external data to the server device 4. The third relay unit 5C also transmits the received external data to the adapter 2 via the network 7 and the access point 3.

[0040] Incidentally, the relay device 5 may not be provided and the adapter 2 and the server device 4 may communicate directly with each other.

[0041] The communication terminal 6 is a mobile terminal device such as a smartphone owned by the user.

[0042] [Adapter configuration] 2 is a block diagram showing a configuration of the adapter 2. As shown in the figure, the adapter 2 has a first communication unit 21, a second communication unit 22, a CPU (Central Processing Unit) 23, and a storage unit 24.

[0043] The first communication unit 21 is a communication IF (Interface) such as a Universal Asynchronous Receiver Transmitter (UART) that is connected by wired communication to the control unit 1B in the indoor unit 1. The second communication unit 22 is a communication IF such as a WLAN that is connected by wireless communication to the access point 3.

[0044] The CPU 23 controls the entire adapter 2. The storage unit 24 has, for example, a Read Only Memory (ROM) and a Random Access Memory (RAM), and stores various types of information such as data and programs.

[0045] The CPU 23 includes an acquisition unit 23A, a transmission unit 23B, a reception unit 23C, an adapter side learning unit 23D, a setting unit 23E, and a prediction unit 23F.

[0046] The acquisition unit 23A acquires sensor data from the indoor unit 1 at a predetermined time interval, for example, every 5 minutes. The acquisition unit 23A stores the sensor data acquired at 5-minute intervals in the sensor data memory 24A. Details of the sensor data will be described later.

[0047] The transmitting unit 23B transmits the sensor data stored in the sensor data memory 24A to the access point 3, the access point 3 transmits the received sensor data to the relay device 5 via the network 7, and the first relay unit 5A of the relay device 5 transmits the received sensor data to the server device 4.

[0048] The receiving unit 23C receives from the server device 4 a learning model (hereinafter also referred to as an AI learning model) generated by the server device 4. In this embodiment, for example, a presence / absence prediction model for predicting the presence / absence of a user is received as the AI ​​learning model, and the received AI learning model is stored in the model memory 24C. The presence / absence prediction model has, for example, a presence / absence pattern indicating the tendency of the presence / absence of a user in the air-conditioned space. For example, the presence / absence pattern has a set of numerical values ​​(learned parameters) indicating the probability of the user's presence at each predetermined time interval in the air-conditioned space.

[0049] The adapter side learning unit 23D updates the learning model used in the adapter 2. Specifically, the adapter side learning unit 23D updates the AI ​​learning model (presence / absence prediction model) stored in the model memory 24C by additional learning based on the sensor data.

[0050] A lot of sensor data (for example, 30 days' worth of sensor data) is required for the server device 44 to generate an AI learning model. Until the adapter 2 receives the AI ​​learning model generated by the server device 4 from the server device 4, the adapter side learning unit 23D determines the initial learning model previously stored in the adapter 2 as the learning model to be used on the adapter 2 side. After receiving the AI ​​learning model from the server device 4, the adapter side learning unit 23D determines the AI ​​learning model as the learning model to be used on the adapter 2 side.

[0051] The adapter side learning unit 23D updates the learned parameters of the initial learning model based on the above sensor data until it receives the AI ​​learning model from the server device 4. After receiving the AI ​​learning model, the adapter side learning unit 23D updates the learned parameters of the AI ​​learning model based on the above sensor data until it next receives the updated AI learning model from the server device 4. In this way, the learning process in which the learning model is additionally learned and the learned parameters are updated based on the sensor data on the air conditioner 100 side rather than on the server device 4 side is called edge learning process.

[0052] The setting unit 23E applies the learning model stored in the model memory 24C to the prediction unit 23F. The prediction unit 23F predicts the presence or absence of a person in the air-conditioned space of the indoor unit 1 based on the learning model.

[0053] Based on the presence / absence prediction result, the control unit 1B of the indoor unit 1 transmits the contents of the air conditioning operation recommended to the user, such as the setting of the air conditioning operation start time or the air conditioning operation stop time, to the user's communication terminal 6. The contents of the recommended air conditioning operation are, for example, the setting of the timer time for starting the air conditioning operation according to the time of getting up, the time of coming home, the time of going out, and the time of going to bed.

[0054] Furthermore, the control unit 1B transmits a signal to the user's communication terminal 6 recommending an additional function operation other than air conditioning operation based on the prediction result of presence or absence. The content of the recommendation of the additional function operation is, for example, a proposal of a timer time for starting the heating sterilization operation at the time when absence is predicted.

[0055] Furthermore, the control unit 1B suspends operation of the air conditioner 100 during the time when it is predicted that no one will be present, based on the result of the prediction of presence or absence.

[0056] Specifically, the control unit 1B extracts time periods (e.g., absence periods of 5 hours or more) during which additional function operation different from normal air conditioning operation can be performed from the 24-hour presence / absence prediction results in the air-conditioned space. The control unit 1B calculates the available operation time from the absence period, compares the calculated available operation time with the operation time of the additional function operation stored in advance, and extracts the time period from the absence period as the time period during which the additional function operation can be performed if the available operation time is equal to or longer than the operation time of the additional function operation stored in advance.

[0057] The storage unit 24 has a sensor data memory 24A, a presence / absence pattern memory 24B, a model memory 24C, a flash memory 24D, an external memory 24E, and a prediction result memory 24F.

[0058] The sensor data memory 24A temporarily stores the sensor data acquired from the indoor unit 1. The presence / absence pattern memory 24B stores the presence / absence pattern of the AI ​​learning model received from the server device 4 and the presence / absence pattern of the initial learning model. The presence / absence pattern is data indicating the tendency of the user's presence / absence in the air-conditioned space. Specifically, it is array data of 1 row and 144 columns (10-minute intervals from 0:00 to 24:00 to 23:50) that takes decimals from 0 indicating absence to 1 indicating presence for each specified time interval.

[0059] The presence / absence pattern of the AI ​​learning model downloaded from the server device 4 is a pattern generated by the server device 4 using, for example, the past 30 days of presence / absence sensor data of the sensor 1C, and the day of the week information and holiday information acquired by the server device 4 from an external server (not shown). In this embodiment, up to five types of presence / absence patterns are generated. FIG. 10 shows the presence / absence patterns of this embodiment. In FIG. 10 and FIG. 11 described later, the presence / absence patterns generated from the sensor data of the air conditioner 100 are shown in the order of the highest presence probability, with the presence probability of the user in the air-conditioned space (for example, the living room) being divided into five stages. The presence / absence patterns are shown in the order of the highest presence probability, namely, presence probability 5, presence probability 4, presence probability 3, presence probability 2, and presence probability 1. The presence / absence patterns are associated with each day of the week so that the CPU 23 can determine which presence / absence pattern the user tends to act in for each day of the week. The association between the day of the week and the presence / absence patterns is performed, for example, by the server device 4.

[0060] For example, as shown in FIG. 10A, Monday and Tuesday correspond to presence / absence pattern 1, as shown in FIG. 10B, Wednesday and Thursday correspond to presence / absence pattern 2, as shown in FIG. 10C, Friday corresponds to presence / absence pattern 3, as shown in FIG. 10D, Saturday corresponds to presence / absence pattern 4, and Sunday corresponds to presence / absence pattern 5, as shown in FIG. 10E.

[0061] The day of the week information is information on the days of the week, Mon. Tues. Wed. Thurs. Fri. Sat. Sun. The holiday information is information that identifies holidays among the days of the week, Mon. Tues. Wed. Thurs. Fri. Sat. Sun. The day of the week information and holiday information are acquired from an external server (not shown) via the second communication unit 22.

[0062] On the other hand, the presence / absence patterns of the initial model are stored in advance regardless of the sensor data of the sensor 1C, and include two patterns: a presence / absence pattern (initial presence / absence pattern 1) that indicates a tendency for people to be outside for a long time (absence time), and a presence / absence pattern (initial presence / absence pattern 2) that indicates a tendency for people to be outside for a short time (absence time). Figure 11 shows a schematic diagram of the initial presence / absence patterns of this embodiment.

[0063] Initial presence / absence pattern 1 is a pattern (weekday pattern) in which the total or continuous absence time in a day (for example, the time when the probability of being absent is equal to or greater than a predetermined value (or the probability of being present is equal to or less than a predetermined value)) is equal to or greater than a predetermined time (for example, 8 hours, hereinafter also referred to as the first absence time). For example, initial presence / absence pattern 1 is a presence / absence pattern as shown in FIG. 11A. Initial presence / absence pattern 2 is a pattern (holiday pattern) in which the total or continuous absence time is less than the first absence time (for example, 5 hours, hereinafter also referred to as the second absence time). For example, initial presence / absence pattern 2 is a presence / absence pattern as shown in FIG. 11B. These presence / absence patterns are, for example, average data of the presence / absence pattern tendency data of a large number of users obtained in advance.

[0064] The model memory 24C stores an AI learning model (first learning model) downloaded from the server device 4, a previous learning model used in the previous presence / absence prediction process, and an online learning model (second learning model) which is used in the user presence / absence prediction process and is a learning model for which additional learning is performed based on sensor data.

[0065] Before the AI ​​learning model is downloaded, the online learning model is the initial learning model itself, or a learning model in which the initial learning model or its learned parameters are additionally learned by the adapter side learning unit 23D based on the sensor data. On the other hand, if the AI ​​learning model has already been downloaded, the online learning model is the downloaded AI learning model itself, or a learning model in which the AI ​​learning model or its learned parameters are additionally learned by the adapter side learning unit 23D based on the sensor data. As will be described later, the AI ​​learning model before additional learning is separately stored for comparison with the newly downloaded AI learning model (determining whether or not they match).

[0066] The flash memory 24D stores the initial learning model used for predicting the presence or absence of the user until the AI ​​learning model is received from the server device 4. When the learning models in the model memory 24C are initialized, they are returned to the initial learning model.

[0067] The initial learning model has a first learned parameter corresponding to initial presence / absence pattern 1 and a second learned parameter corresponding to initial presence / absence pattern 2. The learned parameter is, for example, 1 row and 144 columns of array data (presence / absence data) indicating the presence probability of a user within a predetermined time interval.

[0068] The external memory 24E stores external data such as weather forecast data, etc. The prediction result memory 24F stores the prediction result of the presence or absence of a user in the air-conditioned space, predicted using the AI ​​learning model or the initial learning model.

[0069] [Server device configuration] 3 is a block diagram showing the configuration of the server device 4. As shown in the figure, the server device 4 has a communication unit 41, a CPU 42, and a storage unit 43.

[0070] The communication unit 41 is a communication IF that connects to the relay device 5 by wired communication, and communicates with the adapters 2 of the multiple air conditioners 100 via the relay device 5. The CPU 42 controls the entire server device 4. The storage unit 43 has, for example, an HDD (Hard Disk Drive), ROM, RAM, etc., and stores various information such as data and programs.

[0071] The CPU 42 includes a cloud-side learning unit 42A, a receiving unit 42B, and a transmitting unit 42C.

[0072] The cloud-side learning unit 42A receives each sensor data from the adapter 2 of each air conditioner 100 via the access point 3, the network 7, and the relay device 5. The cloud-side learning unit 42A then generates an AI learning model using the sensor data for a predetermined period (e.g., six days) stored in the sensor data memory 43A among the sensor data received from each adapter 2. Since the cloud-side learning unit 42A generates an AI learning model using all of the sensor data for the predetermined period, it is possible to generate a learning model with high prediction accuracy even when irregular sensor data that should not be learned is included in some of the data. Learning that uses all of the sensor data in this way is sometimes called batch learning.

[0073] The cloud-side learning unit 42A generates or updates an AI learning model for the air conditioner 100 based on the learning results. Then, the cloud-side learning unit 42A stores the newly generated AI learning model or a learning model that is an update of an existing AI learning model in the cloud-side learning memory unit 43B. The cloud-side learning unit 42A can also receive sensor data from multiple air conditioners 100 and generate or update an AI learning model for each air conditioner 100.

[0074] The receiver 42B receives sensor data and the like from each adapter 2 via the access point 3, the network 7, and the relay device 5. The transmitter 42C transmits an AI learning model newly generated for each air conditioner 100 by the cloud-side learning unit 42A or an updated learning model of an existing AI learning model to the adapter 2 of each air conditioner 100 via the relay device 5, the network 7, and the access point 3.

[0075] The storage unit 43 has a sensor data memory 43A and a cloud-side learning storage unit 43B. The sensor data memory 43A stores the sensor data received from the adapter 2 of the air conditioner 100 for each air conditioner 100.

[0076] Each air conditioner 100 has an air conditioner ID for identifying the air conditioner 100, and when the adapter 2 transmits sensor data, it also transmits this air conditioner ID. Then, the server device 4 that receives the sensor data from the adapter 2 stores the sensor data for each air conditioner ID included in the received signal in the sensor data memory 43A. The cloud-side learning storage unit 43B stores the AI ​​learning model generated or updated by the server device 4.

[0077] [System Operation] Next, the operation of the system configured as above will be described.

[0078] (Learning process by server device) First, the learning process in the server device 4 will be described. The operation of the server device 4 is executed by the cooperation of the CPU 42 of the server device 4 and software stored in the storage unit 43. For convenience, in the following description, the CPU 42 is the subject of operation. FIG. 4 is a flowchart showing the flow of the learning process by the server device 4.

[0079] As shown in the figure, the CPU 42 (cloud-side learning unit 42A) of the server device 4 first determines whether or not sensor data has been received from the adapter 2 (step 41). Note that the sensor data transmitted from the adapter 2 is, for example, data for six days.

[0080] If it is determined that the sensor data has been received (Yes in step 41), the CPU 42 stores the sensor data in the sensor data memory 43A (step 42). If it is determined that the sensor data has not been received (No in step 41), the CPU 42 ends the process.

[0081] Next, the CPU 42 judges whether or not there is sensor data stored in the sensor data memory 43A for a predetermined period from the most recent past (step 43). For the presence / absence prediction model in this embodiment, the predetermined period is, for example, about one to two months. If it is judged that there is not sensor data for the predetermined period (No in step 43), the CPU 42 ends the process.

[0082] When it is determined that the sensor data exists for a predetermined period or more (Yes in step 43), the CPU 42 generates an AI learning model that predicts the presence or absence of a person based on the sensor data (step 44).

[0083] Next, the CPU 42 stores the generated AI learning model in the cloud-side learning storage unit 43B to update it (step 45).

[0084] Then, the CPU 42 transmits the AI ​​learning model to the adapter 2 (step 46). The CPU 42 executes the above process every time it receives sensor data from the adapter 2.

[0085] (Air conditioner operation) Next, a description will be given of the operation of the air conditioner 100. The operation of the air conditioner 100 is executed by cooperation between hardware such as the CPU 23 of the adapter 2 of the air conditioner 100 and software stored in the storage unit 24. For convenience, in the following description, the CPU 23 is regarded as the subject of operation.

[0086] 5 is a flowchart showing the flow of the presence / absence prediction process and the learning process by the adapter 2. The learning process is a process in which the CPU 23 updates (additionally learns) the learned parameters of the online learning model being used.

[0087] As shown in the figure, the CPU 23 of the adapter 2 first determines whether or not the AI ​​learning model received from the server device 4 exists in the model memory 24C (step 51).

[0088] When it is determined that the AI ​​learning model received from the server device 4 exists in the model memory 24C (Yes in step 51), the CPU 23 executes the presence / absence prediction process using the AI ​​learning model as an online learning model (step 53).

[0089] On the other hand, if it is determined that the AI ​​learning model received from the server device 4 does not exist in the model memory 24C (No in step 51), the CPU 23 expands the initial learning model stored in the flash memory 24D into the model memory 24C as an online learning model, and executes presence / absence prediction processing (step 52).

[0090] Next, the CPU 23 determines whether or not a predetermined time period of sensor data has been received from the indoor unit 1 (step 54). The predetermined time period is, for example, 30 hours, but is not limited to this.

[0091] When it is determined that the sensor data for the predetermined time period has been received (Yes in step 54), the CPU 23 determines whether or not the sensor data is valid as data to be used in the learning process of the online learning model (step 55).

[0092] Here, the presence / absence sensor data (raw data) indicates, for example, the amount of human activity in an air-conditioned space for five minutes on a five-level scale: 1: no activity (absent), 2: low activity, 3: medium activity, 4: high activity, 5: indefinite activity, and includes a timestamp indicating the time the data was collected. If the CPU 23 determines that the amount of activity for the five minutes has not been updated at a specified time in a day or throughout the day, it determines that the sensor data is invalid.

[0093] Specifically, for example, if the sensor data indicates either low, medium or high activity during a specified period of time or throughout the day, and there is no activity, or if there is no activity during a specified period of time or throughout the day, and there is neither low, medium nor high activity, the sensor data is determined to be invalid.

[0094] If the CPU 23 determines that the sensor data is not valid (No in step 55), the CPU 23 stops the learning process and returns to step 51. If the CPU 23 determines that the sensor data is valid (Yes in step 55), the CPU 23 preprocesses the sensor data for learning (step 56).

[0095] Specifically, the CPU 23 uses the sensor data from 3:00 on the learning day up to 30 hours ago (up to 21:00 on the day before the learning day) as input. The CPU 23 outputs the presence / absence data from 0:00 to 24:00 on the day before the learning day at 10-minute intervals from the input sensor data.

[0096] More specifically, the CPU 23 first converts the five-level activity amount data into presence / absence information indicated by two integers, present (1) / absent (0). For example, if the activity amount for five minutes is small, medium, or large, it is determined to be present (1), and if there is no activity amount or the activity amount is indefinite, it is determined to be absent (0) (for example, the upper table of FIG. 6). If the presence / absence is quantified using the sensor data obtained by converting the five-minute activity amount data into presence / absence information as it is, there is a risk that the tendency of the user's presence / absence in the air-conditioned space cannot be correctly grasped. For example, even if the user is present in the air-conditioned space in which the indoor unit 1 is installed (for example, the living room), the sensor 1C may not detect the user and erroneously determine that the user is absent, or the user may erroneously determine that the user tends to be absent from the air-conditioned space by repeatedly going back and forth between the air-conditioned space and another room (for example, a child's room or a bedroom) for a short time. In such a case, if the interval between the presence and absence in the presence / absence information is within a predetermined time (for example, within 180 minutes), the absence during that time may be complemented with presence. In this embodiment, in order to complement the presence / absence pattern on the day before the learning day, sensor data from the day before the learning day (until 21:00) and sensor data from the learning day itself (until 3:00) are obtained.

[0097] In this way, a group is created from 144 pieces of sensor data, dividing the 24 hours into 10-minute intervals within the range of 0:00 to 24:00 on the day before the learning day, and the presence / absence value (1 / 0) corresponding to that time is entered (for example, the lower table in Figure 6).

[0098] In the pre-processing, the CPU 23 executes a process that takes into consideration the timestamp difference of the input sensor data. FIG 6 is a diagram illustrating the process.

[0099] The input sensor data is data obtained by converting activity data for a predetermined time (for example, 5 minutes) into presence / absence information, and is set to be collected at predetermined time intervals (for example, every 5 minutes) as described above, has a timestamp indicating the collection time, and is accumulated in the order of the timestamp. However, an error may occur in the timestamp. Since the error accumulates, if the time is used as a reference to enter the presence / absence value, in the worst case, an error of about 10 minutes may occur in the timestamp of the sensor data within the range of 0:00 to 24:00 on the day before the learning day. Therefore, when entering the presence / absence value into one group created by 144 pieces of sensor data divided into 10-minute intervals over 24 hours, the CPU 23 does not refer to the time, but instead focuses on the number of pieces of data and enters the 144 pieces of data in the order in which the data was collected.

[0100] On the other hand, if only the number of pieces of data is considered, time errors in the time stamps can be eliminated, but it is not possible to determine data loss that may occur due to, for example, a power outage causing a power outage (since the sensor data from 3:00 on the learning day back to 21:00 on the day before the learning day is used as input, even if several hours of data are missing, the data from before that will be counted.) For this reason, the CPU 23 sets a permissible range for the time lag of the entire data, and if there is an error beyond this range, it determines that there is a data defect (loss) and halts learning.

[0101] That is, as shown in Fig. 6, the CPU 23 determines the data closest to 21:00 on the day before the learning day as the first data, and sets the 360th data as the last data. The CPU 23 judges whether the timestamp of the last data is within a predetermined range of 3:00 on the learning day (whether the timestamp is within a predetermined range around 3:00, which is the timestamp that the 360th acquired sensor data should have when the above setting (5 minute intervals) is followed). If the predetermined range is exceeded, the CPU 23 determines that there is a missing data and suspends learning. In this embodiment, the predetermined range is set to ±10 minutes (2:50 to 3:10).

[0102] With this configuration, even if the timestamp of the sensor data deviates from the set time, it is possible to reliably detect defects (missing parts) in the sensor data. As a result, in this embodiment, it is possible to stop additional learning using a data set in which sensor data is considered to be missing. If there is no missing part in the data set collected after that (for example, the next day's presence / absence data), additional learning is performed using that data set.

[0103] Next, the CPU 23 executes an update (additional learning) process of the presence / absence pattern of the online learning model based on the sensor data (presence / absence data) after the above preprocessing (step 57).

[0104] Specifically, first, the CPU 23 extracts a presence / absence pattern from the current online learning model. As described above, the initial learning model has two presence / absence patterns, but after downloading the AI ​​learning model from the server device 4, it has up to one to five presence / absence patterns.

[0105] As described above, since there are multiple presence / absence patterns (two or more), CPU 23 selects the presence / absence pattern to be updated. Specifically, CPU 23 calculates the cosine similarity of the stored multiple presence / absence patterns and pre-processed sensor data (presence / absence data) as array data of 1 row and 144 columns (10-minute intervals from 0:00 to 24:00 to 23:50). The presence / absence pattern with the highest cosine similarity is selected as the update target. Then, CPU 23 updates the presence / absence pattern in presence / absence pattern memory 24B for the combination of presence / absence pattern (a) and presence / absence data (b) with the highest similarity, using the update formula shown in FIG. 7 (the difference between the presence / absence pattern and the presence / absence data multiplied by a weight parameter).

[0106] That is, CPU23 applies the update equation to each of the above 144 elements of the presence / absence pattern, and overwrites the calculation result (a') calculated using the update equation over the original presence / absence pattern (a). CPU23 overwrites the presence / absence pattern of the online learning model with the presence / absence pattern updated by the overwriting. The above is the update (additional learning) process of the presence / absence pattern of the online learning model. Note that the value of the weight parameter w of the update equation in FIG. 7 is set to 0.1, for example, but is not limited to this.

[0107] Next, the CPU 23 judges whether the sensor data has been stored in the sensor data memory 24A for a predetermined period (step 58). The predetermined period is, for example, six days, but is not limited to this.

[0108] When it is determined that the sensor data has been accumulated for the predetermined period (Yes in step 58), the CPU 23 transmits the sensor data for the predetermined period to the server device 4 via the network 7 and the first relay unit 5A (step 59). The sensor data transmitted here is the presence / absence data before the above-mentioned preprocessing, but it may be the presence / absence data after the above-mentioned preprocessing.

[0109] The CPU 23 repeatedly executes the above process while power is being supplied to the indoor unit 1 and the adapter 2. After the CPU 23 transmits the sensor data for the predetermined period to the server device 4, an AI learning model is generated by the cloud-side learning unit 42A, as described above in the flowchart shown in FIG.

[0110] Next, every time the latest AI learning model is generated by the cloud-side learning unit 42A of the server device 4, the CPU 23 receives the latest AI learning model from the server device 4. Here, we will explain the learning model update process performed on the air conditioner 100 side when there is the latest AI learning model received by the CPU 23 and an online learning model updated by edge learning. Figure 8 is a flowchart showing the flow of the online learning model update process by the adapter 2. Also, Figure 9 is a diagram explaining the update process.

[0111] As shown in FIG. 8, the CPU 23 first determines whether or not an AI learning model has been received from the server device 4 (step 81). In this embodiment, a presence / absence prediction model is exemplified as the AI ​​learning model received from the server device 4. However, the present invention is not limited to this. For example, a prediction model other than the presence / absence prediction model, such as a sensible temperature prediction model that predicts the user's sensible temperature, may also be used. Therefore, when the AI ​​learning model downloaded from the server device 4 includes multiple prediction models (for example, a presence / absence prediction model and various other prediction models), the CPU 23 expands the multiple prediction models downloaded from the server device 4 in the model memory 24C as shown in FIG. 9.

[0112] In this embodiment, the CPU 23 extracts a presence / absence prediction model from the AI ​​learning model (a plurality of various prediction models) deployed in the model memory 24C.

[0113] When the CPU 23 determines that it has received an AI learning model (Yes in step 81), it compares the received AI learning model (presence / absence prediction model) with the previous learning model (previous presence / absence prediction model) stored in the model memory 24C (step 82, Figure 9).

[0114] Next, the CPU 23 judges whether or not the received AI learning model and the previous learning model are inconsistent based on the comparison result (step 83, FIG. 9). The inconsistency between the two models is judged, for example, by whether or not the learned parameters (the presence / absence data) of the two models are completely consistent. However, it may be judged, for example, by whether or not the two parameters are substantially consistent, such as whether the difference between the two parameters is within a predetermined threshold (for example, several percent).

[0115] When it is determined that the presence / absence prediction model and the previous presence / absence prediction model do not match (Yes in step 83), the CPU 23 determines whether or not the received presence / absence prediction model is usable (step 84, FIG. 9).

[0116] If the presence / absence prediction model does not match the previous presence / absence prediction model and is determined to be usable (Yes in step 84), the CPU 23 overwrites and updates the previous presence / absence prediction model and the online learning model in the model memory 24C with the received presence / absence prediction model (step 85, FIG. 9).The CPU 23 then executes presence / absence prediction processing using the overwritten online learning model, and executes processing such as setting the timer time for starting air conditioning operation based on the result, and recommending operation of additional functions or suspension of operation during the predicted absence time.

[0117] On the other hand, if it is determined that the learned parameters of the presence / absence prediction model match the learned parameters of the previous presence / absence prediction model (No in step 83), and if it is determined that the AI ​​learning model is not available (No in step 84), the CPU 23 does not overwrite the previous presence / absence prediction model and the online learning model in the model memory 24C with the received presence / absence prediction model. That is, the CPU 23 maintains the previous presence / absence prediction model and the online learning model in the model memory 24C (step 86). The CPU 23 then continues the presence / absence prediction process using the maintained online learning model, and executes the setting of the timer time for starting the air conditioning operation based on the result, the recommendation process for operating additional functions or pausing operation during the predicted absence time, and the like.

[0118] The CPU 23 repeatedly executes the above process each time it downloads an AI learning model (e.g., a presence / absence prediction model) from the server device 4. With this configuration, it is possible to store a previous learning model (e.g., a previous presence / absence prediction model) for comparison and determine whether the latest AI learning model and the previous learning model match (whether there is a difference). This makes it possible to avoid unnecessary update processing when there is no difference between the latest AI learning model and the previous learning model. Furthermore, when an online learning model has been additionally learned by edge learning processing, it is also possible to avoid losing the additionally learned online learning model.

[0119] [summary] As described above, according to this embodiment, the air conditioner 100 can execute the presence / absence prediction process using the initial learning model until the AI ​​learning model is received from the server device 4. This allows the user of the air conditioner 100 to receive the benefit of the learning model early on (for example, receiving a recommendation to set the timer time for starting air conditioning operation based on the presence / absence prediction result, or receiving a recommendation to operate additional functions or suspend operation during the predicted absence time) even before receiving the AI ​​learning model. Furthermore, even if the initial learning model cannot make an appropriate prediction, edge learning process is performed using sensor data in the air conditioner 100. Therefore, the learned parameters of the initial learning model are updated to appropriate values, and appropriate predictions can be made using the online learning model. Furthermore, even after the AI ​​learning model is received, edge learning process is performed using sensor data on the air conditioner 100 side. Therefore, even if the period from receiving the AI ​​learning model to receiving the next AI learning model becomes long, the learned parameters of the AI ​​learning model are additionally learned by the edge learning process, so that the online learning model is appropriately updated and appropriate predictions (for example, predictions of the presence / absence of the user) can be made using the online learning model.

[0120] [Variations] The present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the spirit and scope of the present invention.

[0121] The data processing and generation intervals such as the intervals for acquiring sensor data, the intervals for transmitting sensor data to the server device 4, and the intervals for learning the learning model shown in the above embodiment are not limited to those described above and can be changed as appropriate. Furthermore, the various threshold values ​​shown in the above embodiment can also be changed as appropriate.

[0122] In the above embodiment, the learning model is an example of a presence / absence prediction model that predicts the presence or absence of people in an air-conditioned space, but the prediction target of the learning model is not limited to presence or absence. For example, the learning model may predict the sensible temperature, temperature unevenness, building heat load, etc., and the above initial learning model and AI learning model are generated and stored accordingly.

[0123] For example, the sensible temperature prediction model is a learning model that predicts the sensible temperature of a user in an air-conditioned space after a predetermined time (e.g., 5 minutes, 10 minutes, etc.) has elapsed, predicts the user's operation preferences according to the predicted sensible temperature, and controls the air conditioner (the set temperature, etc.) based on the operation preferences. In this case, temperature distribution, indoor temperature, set temperature, outdoor temperature, timestamp, cloud cover, etc. are used as sensor data and external data, and learned parameters are generated from them.

[0124] The building heat load prediction model is a learning model that predicts the heat load of a building after a specified time (e.g., one day) has elapsed, and creates and executes an operation plan to operate the air conditioner more efficiently (energy saving) according to the predicted heat load. In this case, the maximum temperature, minimum temperature, average temperature, maximum humidity, minimum humidity, average humidity, rainfall, solar radiation, average air-conditioned space temperature, etc. of the day before the prediction target day, and the predicted maximum temperature, predicted minimum temperature, rainfall, room temperature at the time of prediction, etc. of the day of the prediction target day are used as sensor data and external data, and the learned parameters are generated from them.

[0125] Among the inventions described in the claims of this application, the invention described as an "information processing method" is one in which each step is automatically performed by at least one device such as a computer through information processing by software, and is not performed by a human using a device such as a computer. In other words, the "information processing method" is an information processing method by computer software, and is not a method in which a human operates a calculation tool called a computer. [Explanation of symbols]

[0126] 1…Indoor unit 2…Adapter 4. Server equipment 7. Network 21…1st Communications Department 22…Second Communications Department 23…CPU 24...Storage section 24C…Model memory (RAM) 24D…Flash memory 41…Communications Department 42…CPU 43...Storage section 100…Air conditioner

Claims

1. An air conditioner having an adapter for communicating with a server device, The adapter comprises: A storage unit that stores an initial learning model having learned parameters related to a predetermined prediction process executed by the air conditioner; A communication unit that collects sensor data detected by a sensor included in the air conditioner, transmits the sensor data to the server device, and receives a first learning model generated by the server device based on the sensor data; A control unit that executes the predetermined prediction process using the initial learning model until the first learning model is received, and executes the predetermined prediction process using the first learning model after the first learning model is received. Air conditioner.

2. The air conditioner according to claim 1, The control unit updates the learned parameters of the initial learning model based on the sensor data until the first learning model is received, and updates the learned parameters of the first learning model based on the sensor data after the first learning model is received. Air conditioner.

3. The air conditioner according to claim 1 or 2, The control unit predicts the presence or absence of a user of the air conditioner in an air-conditioned space by using the initial learning model and the first learning model. Air conditioner.

4. The air conditioner according to claim 3, The initial learning model has a plurality of the learned parameters to correspond to a plurality of presence / absence tendencies of the occupant in the air-conditioned space. Air conditioner.

5. The air conditioner according to claim 4, The plurality of learned parameters include a first learned parameter corresponding to a presence / absence pattern having a first absence time as an absence time in the air-conditioned space, and a second learned parameter corresponding to a presence / absence pattern having a second absence time shorter than the first absence time as an absence time in the air-conditioned space. Air conditioner.

6. The air conditioner according to claim 1 or 2, The control unit starts collecting the sensor data when external power is supplied to the air conditioner, and continues collecting the sensor data regardless of whether an air conditioning operation of the air conditioner is started or stopped. Air conditioner.

7. The air conditioner according to claim 1 or 2, Each time the control unit receives the latest first learning model from the server device, the control unit compares the received first learning model with a first learning model received previously, and if the two models do not match, updates the previously received first learning model with the latest first learning model and a second learning model obtained by additionally learning the previously received first learning model based on the sensor data, stores the updated second learning model in the storage unit, and executes the predetermined prediction process with the updated second learning model. Air conditioner.

8. The air conditioner according to claim 2, The sensor data is set to be collected at a predetermined time interval, has a time stamp indicating the time of collection, and is stored in order of the time stamp; The control unit determines whether a timestamp of a predetermined sensor data from among sensor data having a predetermined timestamp is within a predetermined time range before and after a timestamp that the sensor data acquired in the predetermined order should have when the setting is followed, and determines that the sensor data is missing when it is determined that the timestamp is outside the predetermined time range before and after. Air conditioner.

Images