JP7211096B2 - Server device and learning method - Google Patents

Description

The present disclosure relates to a server device and a learning method.

For example, Patent Literature 1 discloses an air conditioning system with a learning function. In addition, in Patent Document 1, when the storage unit has a learning function, control by standard specification settings (for example, automatic driving using the detection result of a human sensor) reflects the preferences and behavior patterns of residents, etc. It is described that a suitable temperature environment is realized for a resident or the like.

JP 2015-117933 A

By the way, a method has also been proposed in which a server collects data related to an air conditioning system and generates a learning model. For example, the server collects user operation information as learning data together with environmental information such as room temperature and outside temperature, and then uses each of these learning data to determine the optimal air conditioner operation when environmental information is input. Generate a learning model that outputs information.

In such a learning model generation method, each learning model has a memory for storing learning data used to generate the learning model. Required learning data is stored and managed in a memory prepared for each learning model (hereinafter referred to as individual memory).

However, the above method has the problem of wasting memory capacity. That is, in the above method, for example, the first learning model learns using three learning data "operating mode, room temperature, outside temperature", and the second learning model learns "operating mode, room temperature, humidity". When three learning data are used for learning, "operating mode, room temperature", which is common learning data for each learning model, is registered in the individual memory for each learning model. In other words, the same data is redundantly stored in the individual memory for each learning model, and the memory capacity is wasted.

In view of such problems, an object of the present disclosure is to provide a server device and a learning method that prevent wasteful consumption of memory capacity.

A server device of one aspect has a receiving unit, a determining unit, and a learning unit. The receiving unit is used for generating a learning model used by a control unit that controls the air conditioner, and receives the learning data from the adapter that collects learning data including data sets of driving history information and operation information for a plurality of learning items. receive data for use and store it in the shared memory. The determination unit is based on a collection pattern that defines a unique learning item for each different learning model used when the control unit that controls the air conditioner generates a learning model, and a learning model to be learned, A unique learning item corresponding to the learning model to be learned is extracted. The learning unit extracts a data set corresponding to the unique learning item extracted by the determination unit from the learning data stored in the shared memory, and based on the extracted data set, the learning model is generated. Generate.

As one aspect, it is possible to prevent wasteful consumption of the capacity of the entire memory.

FIG. 1 is a diagram showing an example of an air conditioning system according to this embodiment. FIG. 2 is a block diagram showing an example of the configuration of the adapter. FIG. 3 is a block diagram showing an example of the configuration of the server device. FIG. 4 is a diagram showing an example of the data structure of the shared memory. FIG. 5 is an explanatory diagram showing an example of the contents of driving history data. FIG. 6 is a diagram showing an example of the data structure of collection patterns. FIG. 7 is a diagram showing an example of the data structure of the model memory. FIG. 8 is a diagram for explaining an example of processing by an updating unit; FIG. 9 is a flow chart showing an example of the processing operation of the server device according to the embodiment.

Hereinafter, embodiments of the server device and the learning method disclosed in the present application will be described in detail based on the drawings. Note that the disclosed technology is not limited by the present embodiment. Further, each embodiment shown below may be modified as appropriate within a range that does not cause contradiction.

FIG. 1 is a diagram showing an example of an air conditioning system according to this embodiment. Air conditioning system 1 shown in FIG. Although indoor units 2A to 2C and adapters 50A to 50C are shown here, the air conditioning system 1 may have other indoor units and adapters.

The indoor unit 2A is, for example, part of an air conditioner that is placed indoors and heats or cools the air in the room. The user of the indoor unit 2A can remotely operate the indoor unit 2A by operating the remote controller 9A. The indoor unit 2A has a main body 3A and a control section 4A that controls the main body 3A. The main body 3A is provided with an indoor fan and an indoor heat exchanger, and the indoor air that has undergone heat exchange with the refrigerant in the indoor heat exchanger is blown out from the main body 3A to heat, cool, dehumidify, etc. the room. will be

The indoor unit 2B has a main body 3B and a control section 4B. Descriptions regarding the indoor unit 2B, the main body 3B, and the control section 4B are the same as those regarding the indoor unit 2A, the main body 3A, and the control section 4A. A user of the indoor unit 2B can remotely operate the indoor unit 2B by operating the remote controller 9B.

The indoor unit 2C has a main body 3C and a controller 4C. Descriptions regarding the indoor unit 2C, the main body 3C, and the control section 4C are the same as those regarding the indoor unit 2A, the main body 3A, and the control section 4A. A user of the indoor unit 2C can remotely operate the indoor unit 2C by operating the remote controller 9C.

The adapter 50A has a communication function that connects the indoor unit 2A and the access point 4 by wireless communication, and a control function that AI (Artificial Intelligence) controls the indoor unit 2A. The adapter 50B has a communication function for connecting the indoor unit 2B and the access point 4 by wireless communication, and a control function for AI-controlling the indoor unit 2B. The adapter 50C has a communication function for connecting the indoor unit 2C and the access point 4 by wireless communication, and a control function for AI-controlling the indoor unit 2C.

In the following description, the indoor units 2A, 2B, and 2C are collectively referred to as the indoor unit 2 unless otherwise distinguished. Also, the adapters 50A, 50B, and 50C are collectively referred to as an adapter 50. FIG.

The access point 4 is, for example, a device that connects the adapter 50A and the communication network 8 by wireless communication using WLAN (Wireless Local Area Network) or the like. The communication network 8 is, for example, a communication network such as the Internet. The communication terminal 7 is a terminal device such as a smartphone owned by a user. The user can use the communication terminal 7 from outside to transmit the operating conditions of the indoor unit 2 (operating mode such as cooling/heating, set temperature, etc.).

The relay device 6 is a device that relays various data related to AI control between the adapter 50 and the server device 100 .

The server device 100 has a function of generating a learning model for the AI that controls the indoor units 2, a database that stores data for learning, and the like. The server device 100 is arranged in, for example, a data center.

FIG. 2 is a block diagram showing an example of the configuration of the adapter. Here, as an example, description will be made using the adapter 50A. As shown in FIG. 2, the adapter 50A has a first communication section 51, a second communication section 52, a storage section 53, and a CPU (Central Processing Unit) . The first communication unit 51 is a communication IF (Interface) such as a UART (Universal Asynchronous Receiver Transmitter) that connects the CPU 54 and the control unit 4A in the indoor unit 2A for communication. The second communication unit 52 is, for example, a communication IF such as WLAN that connects the CPU 54 and the access point 4 for communication. The storage unit 53 has, for example, a ROM (Read Only Memory) and a RAM (Random Access Memory), and stores various information such as data and programs. The CPU 54 controls the adapter 50 as a whole.

The storage unit 53 in the adapter 50A shown in FIG. 2 includes a driving history memory 53A that temporarily stores driving history data acquired from the indoor unit 2A, a model memory 53B that stores learning models acquired from the server device 100, and a weather forecast. and an external memory 53C for storing external data such as.

The CPU 54 has an acquisition section 54A, a transmission section 54B, a reception section 54C, a setting section 54D, and a prediction control section 54E.

The acquisition unit 54A acquires the driving history data, the operation history data, and the time stamp of the acquisition timing from the indoor unit 2A in a predetermined cycle, for example, every 5 minutes. Acquisition unit 54A stores the driving history data, the operation history data, and the time stamp, which are obtained in a 5-minute cycle, in driving history memory 53A.

The transmission unit 54B transmits the driving history data, the operation history data, and the time stamp stored in the driving history memory 53A to the server device 100 via the communication network 8. FIG.

The receiving unit 54C receives the learning model from the server device 100 via the communication network 8, and stores the received learning model in the model memory 53B.

The setting unit 54D applies the learning model stored in the model memory 53B to the predictive control unit 54E.

The predictive control unit 54E controls the control unit 4A in the indoor unit 2A based on the learning model applied by the setting unit 54D. For convenience of explanation, the predictive control unit 54E controls the control unit 4A in the indoor unit 2A based on the learning model. 3A may be controlled directly. Also, the predictive control unit 54E transmits a signal including a control mode based on the learning model to the control unit 4A. That is, the prediction control unit 54E may indirectly control the main body 2A via the control unit 4A, and can be changed as appropriate.

The hardware configuration of the adapters 50B and 50C is the same as that of the adapter 50A. It should be noted that the targets of processing by the adapter 50B are the indoor unit 2B, the main body 3B, and the control unit 4B. The targets of processing by the adapter 50C are the indoor unit 2C, the main body 3C, and the control unit 4C.

FIG. 3 is a block diagram showing an example of the configuration of the server device 100. As shown in FIG. The server device 100 shown in FIG. 3 has a communication unit 110, a storage unit 120, and a CPU . The communication unit 110 is a communication IF that connects the CPU 130 and the relay device 6 for communication. The storage unit 120 has, for example, an HDD (Hard Disk Drive), ROM, RAM, etc., and stores various information such as data and programs. The CPU 130 controls the server device 100 as a whole.

Storage unit 120 in server device 100 shown in FIG. 3 has shared memory 120A, collection pattern memory 120B, and model memory 120C.

Shared memory 120A is a memory that stores driving history data, operation history data, and time stamps transmitted from adapter 50 . FIG. 4 is a diagram showing an example of the data structure of the shared memory. As shown in FIG. 4, this shared memory 120A stores time stamps, driving history data, and operation history data in association with each other. Each set of driving history data and operation history data stored in shared memory 120A is a data set acquired at the time indicated by the time stamp.

FIG. 5 is an explanatory diagram showing an example of the contents of driving history data. The driving history data has a plurality of items, and each item is associated with an item number for uniquely identifying the item. For example, items in the operation history data include operation status, operation mode, set temperature, indoor temperature, indoor humidity, air volume, wind direction, motion sensor, radiation sensor, indoor heat exchanger temperature, outdoor temperature, compressor rotation speed, outdoor Air volume, operating current, outdoor heat exchanger temperature, and the like. Furthermore, the operation history data includes, for example, discharge temperature, compressor temperature, expansion valve opening, radiator temperature, start failure history, abnormal stop history, emergency operation history, air conditioner ID, installation location, facility type, and the like. .

The operating state is the ON-OFF state of the operation of the indoor unit 2 . The operation mode is an operation mode such as cooling or heating of the indoor unit 2 . The set temperature is the indoor target temperature in which the indoor unit 2 is used. The room temperature is the actual temperature in the room where the indoor unit 2 is used. The indoor humidity is the actual humidity in the room where the indoor unit 2 is used. The air volume is the air volume of the conditioned air blown out from the indoor unit 2 . The wind direction is the wind direction of the conditioned air blown out from the indoor unit 2 . The human sensor is the result of detection by the sensor of the presence or absence of people in the room and the amount of activity. The radiation sensor is the result of detection by sensors of the temperature of the floor and walls in the room. The indoor heat exchanger temperature is the temperature of an indoor heat exchanger (not shown) forming part of the body 2A of the indoor unit 2 . The outdoor temperature is the actual outdoor temperature. The compressor rotation speed is the operating rotation speed of a compressor provided in an outdoor unit (not shown) connected to the indoor unit 2 by a refrigerant pipe. The outdoor air volume is the air volume of an outdoor fan provided in the outdoor unit. The operating current is, for example, the operating current of the entire air conditioner such as the indoor unit 2 and the outdoor unit. The outdoor heat exchanger temperature is the temperature of an outdoor heat exchanger provided in the outdoor unit. The discharge temperature is the temperature of the refrigerant discharged from the compressor. Compressor temperature is the temperature at the bottom of the compressor. The expansion valve opening is the opening of an electronic expansion valve provided in the outdoor unit. The radiator temperature is the temperature of the power semiconductor that drives and controls the compressor. The start failure history is a history of compressor start failures. The abnormal stop history is a history of abnormal stop. The emergency operation history is the execution history of emergency operation. The air conditioner ID is an ID given to the indoor unit 2 to identify the air conditioner such as the indoor unit 2 . The installation location is the address of the location where the air conditioner is installed. The facility type is the type of facility such as a store, restaurant, or factory where the air conditioner such as the indoor unit 2 is installed.

The operation history data includes data used for home air conditioners and data used for commercial air conditioners. Operational history data used in home air conditioners includes, for example, operation status, operation mode, set temperature, indoor temperature, indoor humidity, air volume, wind direction, values detected by human sensors, and values detected by radiation sensors. They are a value, a time stamp, an air conditioner ID, an installation location, and the like. Air conditioners for home use use AI to operate and make proposals in pursuit of comfort and energy saving, so for example, set temperature, operation mode, indoor and ambient environment, etc. are necessary data for home use.

Operation history data used for commercial air conditioners includes, for example, operation status, operation mode, set temperature, indoor temperature, indoor humidity, air volume, wind direction, values detected by human sensors, and values detected by radiation sensors. These are the detected values, the indoor heat exchanger temperature, the outdoor temperature, the compressor rotation speed, the outdoor air volume, the operating current, the outdoor heat exchanger temperature, and the like. Further, other operation history data includes, for example, discharge temperature, compressor temperature, expansion valve opening, radiator temperature, start failure history, abnormal stop history, emergency operation history, time stamp, air conditioner ID, installation location, Facility type, etc. In commercial air conditioners, AI predicts failures and the need for maintenance of each device. is determined by AI. In addition to the above, operational history data necessary for commercial air conditioners is, for example, data used for failure prediction. In addition, the rotation speed of the compressor and fan motor used in the air conditioner are stopped while the air conditioner is stopped, and each operation history data is not generated. and data such as outdoor heat exchanger temperature shall not be obtained while the air conditioner is stopped.

The operation history data indicates the user's operation history for the indoor unit 2 . For example, the operation history data includes changed temperature setting, air volume, air velocity, and the like.

The collected pattern memory 120B is a memory that stores a collected pattern 120BB that associates a learning model with driving history data used to generate the learning model. FIG. 6 is a diagram showing an example of the data structure of the collection pattern memory. As shown in FIG. 6, in the collection pattern 120BB, learning model identification information is associated with item numbers. The learning model identification information is information that uniquely identifies the generated learning model. The learning target item number is information that uniquely identifies an item of the driving history data used when generating the learning model. For example, an item of driving history data used when generating a learning model is an example of a "learning item."

For example, the learning model with the learning model identification information “E001” indicates that it is generated using data of item numbers “1 to 7” among all the items described in FIG. It indicates that the learning model with the learning model identification information “E002” is generated using the data of the items with item numbers “1 to 11”. The learning model with learning model identification information “E003” indicates that it is generated using the data of item numbers “1 to 15” out of all the items described in FIG.

The model memory 120C stores a learning model learned by a learning section 130C, which will be described later. FIG. 7 is a diagram showing an example of the data structure of the model memory. As shown in FIG. 7, the model memory 120C stores learning model identification information and learning models in association with each other.

As shown in FIG. 3, the CPU 130 of the server device 100 has a receiving section 130A, a determining section 130B, a learning section 130C, an updating section 130D, and a notification section 130E.

The receiving unit 130A receives the driving history data, the operation history data, and the time stamp from the adapter 50, associates the received driving history data, the operation history data, and the time stamp, and stores them in the shared memory 120A. Store. Receiving unit 130</b>A repeatedly executes the above process every time driving history data, operation history data, and a time stamp are received from adapter 50 .

When generating a learning model, the determination unit 130B extracts an item number corresponding to the learning model identification information from the collection pattern 120BB stored in the collection pattern memory 120B according to the learning model identification information of the learning model to be generated. . The determination unit 130B outputs the learning model identification information and the determined item number to the learning unit 130C. In the following description, the learning model identification information of the learning model to be learned is appropriately referred to as "learning target identification information".

For example, when determination unit 130B receives learning object identification information “E001”, it extracts “item numbers 1 to 7” as learning object item numbers. In this case, the determination unit 130B outputs the learning target identification information “E001” and the learning target item numbers “item numbers 1 to 7” to the learning unit 130C.

Here, the process of designating the learning target identification information to the determination unit 130B may be performed in any manner. For example, the administrator of the server device 100 may operate the input device to specify the learning target identification information, or the determination unit 130B may determine whether a predetermined period of time has elapsed ( For example, the learning model identification information of a learning model that has passed 48 hours) may be automatically selected as the learning target identification information.

The learning unit 130C acquires the learning target identification information and the learning target item number, and stores the driving history data of the item corresponding to the learning target item number, the corresponding operation history data, and the time stamp in the shared memory. 120A. For example, learning unit 130C acquires driving history data and operation history data for 48 hours. The learning unit 130C generates a learning model corresponding to the learning target identification information based on the data of the learning target item and the operation history data. The learning unit 130C associates the generated learning model with the learning target identification information (learning model identification information) and stores them in the model memory 120C.

Here, when a collection pattern update request is received, the update unit 130D, which will be described later, executes the processing described below, so that even when an item corresponding to the learning model is changed, the learning model can be properly learned, can be updated.

The update unit 130D updates the item number to be learned when receiving update request information for the item number to be learned stored in the collection pattern 120BB. The update request information includes a set of learning model identification information to be updated and a post-update item number.

FIG. 8 is a diagram for explaining an example of processing by an updating unit; In the example shown in FIG. 8, it is assumed that the learning model identification information “E001” and the item number “item numbers 1 to 7, 11, 12” of the learning object are set in the update request information 135 . In other words, item numbers 11 and 12 are added to the learning object item numbers of the learning model identification information “E001”. When updating unit 130D receives update request information 135, update unit 130D replaces learning target item numbers “1 to 7” corresponding to learning model identification information “E001” of collection pattern 120BB with learning target item numbers “1 to 7”. , 11, 12”.

Here, when the collection pattern 120BB is updated, the determination unit 130B determines the update time, the updated item number of the learning object, and the learning model identification information corresponding to the updated item number of the learning object. , to the learning unit 130C. In the following description, the learning model identification information corresponding to the updated item number of the learning target is appropriately referred to as "updated learning model identification information".

When learning unit 130C acquires the update time, the updated learning target item number, and the updated learning model identification information, learning unit 130C obtains the learning target item numbers up to a predetermined time (for example, 48 hours) before the update time. and the operation history data corresponding thereto from the shared memory 120A. Learning unit 130C performs updating based on the driving history data of the item corresponding to the item number to be learned and the operation history data corresponding thereto, up to a predetermined time (for example, 48 hours) before the update time. A learning model corresponding to the learning model identification information is generated. The learning unit 130C associates the generated learning model with updated learning model identification information (learning model identification information) and stores them in the model memory 120C.

Here, the process of designating the update request information to the update unit 130D may be performed in any manner. For example, the administrator of the server device 100 may operate the input device to input the update request information, or the user may use the communication terminal 7 to notify the server device 100 of the update request information.

The notification unit 130E is a processing unit that notifies the adapter 50 of each learning model stored in the model memory 120C.

Next, the operation of the server device 100 according to this embodiment will be described. FIG. 9 is a flow chart showing an example of the processing operation of the server device according to the embodiment. As shown in FIG. 9, the receiving unit 130A of the server device 100 receives driving history data, operation history data, and time stamps from the adapter 50 (step S101).

The receiver 130A stores the driving history data, the operation history data, and the time stamp in the shared memory 120A (step S102). When generating a learning model, the determination unit 130B of the server device 100 compares the learning target identification information of the learning model to be generated with the collection pattern 120BB (step S103). The determination unit 130B extracts the item number corresponding to the learning target identification information (step S104).

The learning unit 130C acquires the driving history data and the operation history data of the item corresponding to the extracted item number from the shared memory 120A (step S105). The learning unit 130C generates a learning model based on the driving history data and the operation history data of the item corresponding to the determined item number (step S106). The learning unit 130C stores the learning model in the model memory 120C (step S107).

The notification unit 130E of the server device 100 notifies the learning model stored in the model memory 120C to the adapter 50 (step S108). If the server device 100 does not end the learning (step S109, No), it proceeds to step S110, and if it ends the learning (step S109, Yes), it ends the processing.

The determination unit 130B of the server device 100 compares the learning target identification information of the next learning model with the collection pattern 120BB (step S110), and proceeds to step S104.

As described above, the server device 100 according to the present embodiment stores the driving history data and the operation history data received from the adapter 50 in the shared memory 120A, and selects one of the plurality of learning models. When generating, the driving history data and operation history data of specific items corresponding to the learning model to be learned are acquired from the shared memory 120A to generate the learning model. In this way, all the driving history data and operation history data are stored in the shared memory 120A, and the driving history data and operation history data necessary for generation are acquired for each learning model. Unlike the case where the operation history data are separately stored in the storage unit 120, duplication of the operation history data and the operation history data does not occur, and wasteful consumption of the memory capacity can be prevented.

Further, when the server device 100 according to the present embodiment receives collection pattern update request information, it updates the collection pattern 120BB (adds or deletes the item number to be learned), and based on the updated item number, Then, the processing for generating the corresponding learning model is executed. A learning model can be appropriately generated and updated simply by adding or deleting item numbers to be learned.

By the way, the case where the adapter 50 transmits the driving history data of the indoor unit 2 to the server device 100 via the relay device 6 has been exemplified, but the driving history data etc. 100, and can be changed as appropriate.

Also, each constituent element of each part illustrated does not necessarily need to be physically configured as illustrated. In other words, the specific form of distribution and integration of each part is not limited to the one shown in the figure, and all or part of it can be functionally or physically distributed and integrated in arbitrary units according to various loads and usage conditions. can be configured as

Furthermore, the various processing functions performed by each device are implemented on a CPU (Central Processing Unit) (or a microcomputer such as an MPU (Micro Processing Unit) or MCU (Micro Controller Unit)), in whole or in part. You can make it run. In addition, various processing functions may be executed in whole or in part on a program analyzed and executed by a CPU (or a microcomputer such as an MPU or MCU) or on hardware based on wired logic. Needless to say.

1 air conditioning system 2A, 2B, 2C indoor unit 3A, 3B, 3C main body 4A, 4B, 4C control unit 50A, 50B, 50C adapter 51 first communication unit 52 second communication unit 53, 120 storage unit 53A operation history Memory 53B Model memory 53C External memory 54, 130 CPU
54A acquisition unit 54B transmission unit 54C, 130A reception unit 54D setting unit 54E prediction control unit 100 server device 120A shared memory 120B collection pattern memory 130B determination unit 130C learning unit 130D update unit 130E notification unit

Claims (4)

The learning data is received from an adapter that is used to generate a learning model used by a control unit that controls an air conditioner and that collects learning data including data sets of driving history information and operation information for a plurality of learning items. and a receiving unit for storing in a shared memory;
Unique learning corresponding to the learning model to be learned based on a collection pattern defining unique learning items used when generating the learning model for each different learning model and the learning model to be learned a determination unit that extracts items;
A learning unit that extracts a data set corresponding to the unique learning item extracted by the determination unit from the learning data stored in the shared memory, and generates the learning model based on the extracted data set. A server device characterized by comprising: When receiving update data in which a new learning item is added to the unique learning item corresponding to the first learning model among the plurality of learning models, the learning item corresponding to the first learning model of the collection pattern includes the 2. The server device according to claim 1, further comprising an updating unit that adds a new learning item and updates the collection pattern. When the collection pattern is updated, the learning unit provides a data set corresponding to the unique learning items corresponding to the first learning model, collected by the adapter during a predetermined period of time before the update. 3. The server device according to claim 2, wherein the first learning model is generated again based on the data set corresponding to the new learning item and the data set corresponding to the new learning item. the computer
The learning data is acquired from an adapter that is used to generate a learning model used by a control unit that controls an air conditioner and that collects learning data including data sets of driving history information and operation information for a plurality of learning items. and storing in shared memory;
Unique learning corresponding to the learning model to be learned based on a collection pattern in which unique learning items used when generating the learning model are defined for each different learning model and the learning model to be learned extracting an item;
extracting a data set corresponding to the extracted unique learning item from the learning data stored in the shared memory, and generating the learning model based on the extracted data set; Characterized learning method.

Images