US20190101305A1

US20190101305A1 - Air conditioning control system

Info

Publication number: US20190101305A1
Application number: US16/144,509
Authority: US
Inventors: Keita HADA; Kazunori Iijima
Original assignee: Fanuc Corp
Current assignee: Fanuc Corp
Priority date: 2017-10-04
Filing date: 2018-09-27
Publication date: 2019-04-04
Also published as: JP2019066135A; DE102018007640A1; CN109612037A

Abstract

An air conditioning control system for controlling an air conditioner specifies a condition in an environment, detects a state amount representing the environment, infers a control method for the air conditioner in the environment based on the state amount that is detected, and controls the air conditioner based on the control method that is inferred. Further, the air conditioning control system generates or updates a plurality of learning models through machine learning using the state amount that is detected and stores the plurality of learning models in a manner such that they are associated with combinations of conditions in the environment.

Description

BACKGROUND OF THE INVENTION

1. Field of the Invention

The present invention relates to an air conditioning control system and especially relates to an air conditioning control system that controls air conditioning while switching learning models in accordance with an installation state and an operation state of machines in a factory.

2. Description of the Related Art

Since a plurality of machines (machine tools, for example) ate installed in a factory and the temperature in the factory affects machining accuracy in precision machining, an air conditioner for controlling environment conditions inside the factory is provided. Production in a factory is instructed based on a production plan. In order to produce products instructed in the production plan by an instructed delivery date, operation states and machining conditions of a plurality of machines arc controlled find an operation state of an air conditioner is controlled to realize temperature conditions (temperature and uniformity) in the factory which are required for machining.
As a technique for controlling air conditioning so as to satisfy temperature conditions in a factory which are required for machining, Japanese Patent Application Laid-Open No. 06-307703 and Japanese Patent Application Laid-Open No. 2015-2063519, for example, disclose a technique in which a plurality of air conditioners are controlled in a lump so as to maintain a temperature in a factory at a predetermined temperature.
However, temperature distribution in a factory varies depending on installation states of machines (machine tools, robots, and the like) as heat generating sources installed in the factory and operation states of respective machines. Therefore, in order to realize temperature conditions in a factory required for machining, it is necessary to control air conditioners while taking into account installation states and operation states of respective machines. It is, however, difficult to appropriately control air conditioning in accordance with such various situations.
It is conceivable to introduce a machine learning device so as to control air conditioning. However, much state information detected in various situations are required and many parameters including data related to situations are required so as to produce a versatile machine learning device (versatile learning model) capable of coping with various situations described above. Accordingly, known problems such as over learning may occur.

SUMMARY OF THE INVENTION

An object of the present invention is to provide an air conditioning control system that is capable of performing appropriate air conditioning control, in which installation states and operation states of machines are taken into account, in a wider range.
An air conditioning control system according to the present invention is provided with a mechanism that switches learning models, which are used for determining an air conditioning control action to be used, depending on installation states and operation states of machines in a factory, thereby solving the above-mentioned problems. The air conditioning control system according to the present invention has a plurality of learning models, selects a learning model depending on installation suites and operation states, for example, of machines in a factory, performs machine learning with respect to the selected learning model based on a state amount detected from the factory, and properly uses learning models thus created, depending on the installation states and the operation states, for example, of machines in the factory so as to control an air conditioner.
An air conditioning control system according to an aspect of the present invention controls an air conditioner in an environment in which at least one machine is installed. The air conditioning control system includes: a condition specification unit that specifies a condition in the environment; a state amount detection unit that detects a state amount representing a state of the environment; an inference calculation unit that infers a control method for the air conditioner in the environment based on the state amount; an air conditioning control unit that controls the air conditioner based on the control method that is inferred by the inference calculation unit; a learning model generation unit that generates or updates a learning model through machine learning using the state amount; and a learning model storage unit that stores one or more learning models generated by the learning model generation unit in a manner such that the one or more learning model are associated with a combination of conditions specified by the condition specification unit. The inference calculation unit calculates a control method for the air conditioner in the environment managed by the air conditioning control system, by selectively using one or more learning models among learning models stored in the learning model storage unit, based on the condition in the environment that is specified by the condition specification unit.
The air conditioning control system may further include a feature amount creation unit that creates a feature amount characterizing the environment based on the state amount detected by the state amount detection unit, wherein the inference calculation unit may infer a control method for the air conditioner in the environment based on the feature amount, and the learning model generation unit may generate or update a learning model through machine learning using the feature amount.
The learning model generation unit may alter an existing learning model stored in the learning model storage unit so as to generate a new learning model.
The learning model storage unit may encrypt and store a learning model generated by the learning model generation unit, and decrypt the encrypted learning model when the learning model encrypted is read by the inference calculation unit.
An air conditioning control system according to another aspect of the present invention controls an air conditioner in an environment in which one or more machines are installed. The air conditioning control system includes: a condition specification unit that specifies a condition in the environment; a state amount detection unit that detects a state amount representing the environment:
an inference calculation unit that infers a control method for the air conditioner in the environment based on the state amount; an air conditioning control unit that controls the air conditioner based on the control method that is inferred by the inference calculation unit; and a learning model storage unit that stores at least one learning model that is preliminarily associated with a combination of conditions in the environment. The inference calculation unit calculates a control method for the air conditioner in the environment by selectively using one or more learning models among the learning models stored in the learning model storage unit, based on the condition in the environment that is specified by the condition specification unit.
The air conditioning control system may further includes a feature amount creation unit that creates a feature amount characterizing the environment based on the state amount, wherein the inference calculation unit may infers, based on the feature amount, a control method for the air conditioner in the environment managed by the air conditioning control system.
An air conditioning controller according to still another aspect of the present invention includes the condition specification unit and the state amount detection unit which are described above.
An air conditioning control method according to yet another aspect of the present invention includes: a step for specifying a condition for controlling an air conditioner in an environment in which one or more machines are installed; a step for detecting a state amount representing the environment; a step for inferring a control method for the air conditioner in the environment based on the state amount; a step for controlling the air conditioner based on the control method; and a step for generating or updating a learning model through machine learning using the state amount. In the step for inferring, a learning model to be used based on the condition in the environment that is specified in the step for specifying is selected from the one or more learning models that are preliminarily associated with a combination of conditions in the environment, and a control method for the air conditioner in the environment is calculated using the learning model that is selected.
The air conditioning control method may further include a step for creating a feature amount characterizing the environment based on the state amount, wherein in the step for inferring, a control method for the air conditioner in the environment may be inferred based on the feature amount, and in the step for generating or updating a learning model, a learning model may be generated or updated through machine learning using the feature amount.
An air conditioning control method according to yet another aspect of the present invention includes: a step for specifying a condition for controlling an air conditioner in an environment in which one or more machines are installed; a step for detecting a state amount representing the environment; a step for inferring a control method for the air conditioner in the environment based on the state amount; and a step for controlling the air conditioner based on the control method. In the step for inferring, a learning model to be used based on the condition in the environment that is specified in the step for specifying is selected from one or more learning models that are preliminarily associated with a combination of conditions in the environment and a control method for the air conditioner in the environment is calculated using the learning model that is selected.
The air conditioning control method may further includes a step for creating a feature amount characterizing the environment based on the state amount, wherein in the step for inferring, a control method for the air conditioner in the environment may be inferred based on the feature amount.
A learning model set according to yet another aspect of the present invention includes a plurality of learning models each of which is associated with a combination of conditions for controlling an air conditioner in an environment in which one or more machines are installed. Each of the plurality of learning models is a learning model generated or updated, in a condition in the environment, based on a state amount, representing the environment, and one learning model is selected from the plurality of learning models based on a condition set in an environment, and the learning model that is selected is used for processing of inferring a control method for the air conditioner in the environment.
According to the present invention, since machine learning can be performed, based on a state amount detected in each state, with respect to a learning model selected depending on installation states and operation states of machines in a factory, machine learning can be efficiently performed while preventing over learning. Further, since an air conditioner is controlled by using a learning model selected depending on installation states and operation states, for example, of machines in a factory, accuracy in control of the air conditioner is improved.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a functional block diagram schematically illustrating an air conditioning control system according to a first embodiment.

FIG. 2 is a drawing showing an example of a model in an environment managed by the air conditioning control system.

FIG. 3 is a functional block diagram schematically illustrating on air conditioning control system according to a second embodiment.

FIG. 4 is a functional block diagram schematically illustrating an air conditioning control system according to a third embodiment.

FIG. 5 is a functional block diagram schematically illustrating an air conditioning control system according to a fourth embodiment.

FIG. 6 is a functional block diagram schematically illustrating an air conditioning control system according to a fifth embodiment.

FIG. 7 is a functional block diagram schematically illustrating a modification of the air conditioning control system according to the fifth embodiment.

FIG. 8 is a functional block diagram schematically illustrating an air-conditioning control system according to a sixth embodiment.

FIG. 9 is a schematic flowchart of processing executed in the air conditioning control system illustrated in any of FIGS. 6 to 8.

FIG. 10 is a schematic flowchart of processing executed in the air conditioning control system illustrated in any of FIG. 1 and FIGS. 3 to 5.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

FIG. 1 is a functional block diagram schematically illustrating an air conditioning control system 1 according to a first embodiment.
Each of functional blocks illustrated in FIG. 1 is implemented such that a processor such as a CPU and a GPU included in a computer such as a numerical controller, a cell computer, a host computer, and a cloud server controls an operation of each unit of a device in accordance with each system program.
An air conditioning control system 1 according to the present embodiment includes an environment management unit 100, an inference processing unit 200, and a learning model storage unit 300. The environment management unit 100 manages an environment (a room in which an air conditioner 130 is installed, for example) at least a state of which is an object of observation and inference. The inference processing unit 200 infers a state of the environment. The learning model storage unit 300 stores and manages a plurality of learning models. This air conditioning control system 1 further includes an air conditioning control unit 400 and a learning model generation unit 500. The air conditioning control unit 400 controls air conditioning based on a result obtained through inference of a state of an environment performed by the inference processing unit 200. The learning model generation unit 500 generates and updates learning models to be stored in the learning model storage unit 300.
The environment management unit 100 according to the present embodiment specifics conditions for controlling the air conditioner 130 in an environment in which the air conditioner 130 controlled by the air conditioning control system 1 is installed (an environment managed by the air conditioning control system 1) and acquires a state amount representing a state of the environment. This environment management unit 100 can be mounted on a central management device of air conditioners or a numerical controller installed in an environment managed by the air conditioning control system 1, for example. The environment management unit 100 is configured to be connected with a wired/wireless network 150 so as to be able to exchange data with machines 120, which are installed in an environment managed by the air conditioning control system 1, and the air conditioners 130 controlled by the air conditioning control system 1.
A condition specification unit 110 included in the environment management unit 100 specifies conditions (installation states of the machines 120 and the air conditioners 130 and operation states of the machines 120, for example) in an environment managed by the air conditioning control system 1.
The installation states of the machines 120 and the air conditioners 130 are represented as how the machines 120 as heat sources are installed in each of a plurality of areas which are obtained by dividing on environment managed by the air conditioning control system 1, as illustrated in FIG. 2, for example. As the installation states of the machines 120 and the air conditioners 130 in the example illustrated in FIG. 2, the machines 120 as heat sources are installed in areas 2, 7, 11, and 12 and the air conditioner 130 is installed in area 5, for example. The environment management unit 100 may acquire such installation states of the machines 120 and the air conditioners 130 through setting by an operator, for example, via an input/output device which is not illustrated.
Operation states of the machines 120 are acquired as operation states of respective machines 120 operating in an environment managed by the air conditioning control system 1. Operation states of the machines 120 may be represented in classification of calorific values corresponding to operations of the machines 120, such as stop, low operation (low temperature), medium operation (medium temperature), and high operation (high temperature), for example. The environment management unit 100 may set operation states of the machines 120 based on operation states of the machines 120 (an internal temperature of the machine, motions of a main spindle, a feed axis, and the like, and a load state, for example) acquired from respective machines 120 via the network 150.
The condition specification unit 110 specifies (outputs) conditions in an environment managed by the air conditioning control system 1, which are acquired as described above, with respect to the learning model storage unit 300 and the learning model generation unit 500. The condition specification unit 110 plays a role in notifying each unit of the air conditioning control system 1 of conditions of current air conditioning control of the environment management unit 100 as conditions for selecting a learning model.
A state amount detection unit 140 included in the environment management unit 100 detects a state of an environment managed by the air conditioning control system 1 as a state amount. Examples of the state amount of an environment managed by the air conditioning control system 1 include an ambient temperature, an internal temperature, and an operation state of each machine 120, and a temperature of each area in the environment managed by the air conditioning control system 1. The state amount detection unit 140 detects detection values which are detected by temperature sensors provided to the machines 120 and the air conditioners 130 and a temperature sensor installed in each area in an environment managed by the air conditioning control system 1, as state amounts. The state amounts detected by the state amount detection unit 140 are outputted to the inference processing unit 200 and the learning model generation unit 500.
The inference processing unit 200 according to the present embodiment observes a state, which is acquired from the environment management unit 100, of an environment managed by the air conditioning control system 1 and infers the environment managed by the air conditioning control system 1 based on this observation result. The inference processing unit 200 can be mounted on an air conditioning controller, a cell computer, a host computer, a cloud server, a machine learning device, or the like, for example.
A feature amount creation unit 210 included in the inference processing unit 200 creates a feature amount representing a feature of an environment managed by the air conditioning control system 1 based on state amounts detected by the state amount detection unit 140. The feature amount which is created by the feature amount creation unit 210 and represents a feature of an environment managed by the air conditioning control system 1 is useful information as a determination material of air conditioning control by the air conditioning control unit 400. Further, the feature amount which is created by the feature amount creation unit 210 and represents a feature of an environment managed by the air conditioning control system 1 is input data to be used when an inference calculation unit 220, which will be described later, performs inference using a learning model. The feature amount which is created by the feature amount creation unit 210 and represents a feature of an environment managed by the air conditioning control system 1 may be an ambient temperature obtained by sampling ambient temperatures, detected by the state amount detection unit 140, of each machine 120 in predetermined sampling cycles for the past predetermined period, a peak value in operation states, detected by the state amount detection unit 140, of each machine 120 in the past predetermined period, or a combination of signal processing with respect to temperatures, detected by the state amount detection unit 140, of respective areas such as integration and conversion into a time-series frequency domain, standardization of an amplitude or power density, adaptation to a transfer function, dimensional reduction to specific time or specific frequency width, for example. The feature amount creation unit 210 performs preprocessing of state amounts detected by the state amount detection unit 140 and normalizes the state amounts so that the inference calculation unit 220 can deal with the state amounts.
The inference calculation unit 220 included in the inference processing unit 200 infers a method for controlling the air conditioner 130 in an environment managed by the air conditioning control system 1 based on a learning model, which is selected from the learning model storage unit 300 based on conditions in a current environment managed by the air conditioning control system 1, and based on a feature amount, which is created by the feature amount creation unit 210. The inference calculation unit 220 is implemented by applying a learning model stored in the learning model storage unit 300 to a platform in which inference processing based on machine learning can be executed. The inference calculation unit 220 may be used for performing inference processing using a multilayer neural network, or used for performing inference processing using a known learning algorithm as machine learning for Bayesian network, a support vector machine, mixture Gaussian model, and the like, for example. The inference calculation unit 220 may be used for performing inference processing using a learning algorithm for supervised learning, unsupervised learning, reinforcement learning, and the like, for example. Further, the inference calculation unit 220 may be capable of executing inference processing respectively based on a plurality of kinds of learning algorithms. The inference calculation unit 220 constitutes a machine learning device based on a learning model, which is selected from the learning model storage unit 300, of machine learning, and executes inference processing using a feature amount created by the feature amount creation unit 210 as input data of this machine learning device so as to infer a method for controlling the air conditioner 130 in an environment managed by the air conditioning control system 1. The method for controlling the air conditioner 130 which is a result of the inference by the inference calculation unit 220 may be a set temperature or a wind direction (including a swing action, for example) of each air conditioner 130 installed in an environment managed by the air conditioning control system 1, for example.
The learning model storage unit 300 according to the present embodiment is capable of storing a plurality of learning models which are associated with combinations of conditions, which are specified by the condition specification unit 110, in an environment managed by the air conditioning control system 1. The learning model storage unit 300 can be mounted on the environment management unit 100, a cell computer, a host computer, a cloud server, a database server, or the like, for example.
The learning model storage unit 300 stores a plurality of learning models 1, 2, . . . , N which are associated with combinations of conditions (installation states and operation states of machines in a factory, for example), which are specified by the condition specification unit 110, in an environment managed by the air conditioning control system 1. The combinations of conditions (installation states and operation states of machines in a factory, for example) in an environment managed by the air conditioning control system 1 here represent combinations of values which can be taken by each condition, ranges of values, and lists of values. In the case where an environment managed by the air conditioning control system 1 is modeled as illustrated in FIG. 2, for example, a matrix including states of respective areas as elements [nothing, machine (high temperature), nothing, nothing, air conditioner, nothing, machine (medium temperature), nothing, nothing, nothing, machine (medium temperature), machine (stop)] may be used as one of combinations of conditions in an environment managed by the air conditioning control system 1.
Learning models stored in the learning model storage unit 300 are stored as information which can constitute one learning model complying with inference processing in the inference calculation unit 220. Leaning models stored in the learning model storage unit 300 may be stored as the number of neurons (perception) of each layer or a weight parameter among neurons (perceptron) of each layer, for example, in the case of learning models using a learning algorithm of the multilayer neural network. Further, learning models stored in the learning model storage unit 300 may be stored as transition probability between a node and a node constituting the Bayesian network, for example, in the case of learning models using a learning algorithm of the Bayesian network. The learning models stored in the learning model storage unit 300 may each be learning models using the same learning algorithm or may each be learning models using different learning algorithms. Thus, each of the learning models stored in the learning model storage unit 300 may be a learning model using any learning algorithm as long as the learning models are applicable to inference processing by the inference calculation unit 220.
The learning model storage unit 300 may store one learning model in a manner such that it is associated with one combination of conditions in an environment managed by a single air conditioning control system 1 or may store two or more learning models using different learning algorithms in a manner such that they are associated with one combination of conditions in an environment managed by a single air conditioning control system 1. The learning model storage unit 300 may store learning models using different learning algorithms in a manner such that they are each associated with a plurality of combinations, ranges of which are overlapped with each other, of conditions in an environment managed by the air conditioning control system 1. In this case, the learning model storage unit 300 further sets use conditions such as required throughput and kinds of learning algorithms with respect to learning models corresponding to combinations of conditions in an environment managed by the air conditioning control system 1. Accordingly, it becomes possible to select learning models corresponding to the inference calculation units 220, whose executable inference processing and throughput are different, with respect to combinations of conditions in an environment managed by the air conditioning control system 1, for example.
When the learning model storage unit 300 receives a reading/writing request of a learning model including a combination of conditions in an environment managed by the air conditioning control system 1 from the outside, the learning model storage unit 300 performs reading/writing with respect to a learning model stored in a manner to be associated with the combination of conditions in the environment managed by the air conditioning control system 1. At this time, the learning model reading/writing request may include information on inference processing which can be executed by the inference calculation unit 220 and throughput of the inference calculation unit 220. In such case, the learning model storage unit 300 performs reading/writing with inspect to a learning model which is associated with a combination of conditions in the environment managed by the air conditioning control system 1, inference processing which can be executed by the inference calculation units 220, and throughput of the inference calculation units 220. The learning model storage unit 300 may have a function of performing reading/writing with respect to a learning model which is associated with (a combination of) conditions, which are specified by the condition specification unit 110, based on these conditions, in response to a reading/writing request of a learning model from the outside. Provision of such function eliminates necessity of providing a function of requiring a learning model based on conditions, which are specified by the condition specification unit 110, with respect to the inference calculation unit 220 and the learning model generation unit 500.
Here, the learning model storage unit 300 may encrypt a learning model generated by the learning model generation unit 500 and store the encrypted learning model, and may decrypt the encrypted learning model when the learning model is read by the inference calculation unit 220.
The air conditioning control unit 400 controls an operation of the air conditioner 130 which is installed in an environment managed by the air conditioning control system 1, based on a method for controlling the air conditioner 130, which is inferred by the inference processing unit 200, in the environment managed by the air conditioning control system 1. This air conditioning control unit 400 transmits a control command to each air conditioner 130 via the network 150, for example, so as to control each air conditioner 130. Further, this air conditioning control unit 400 may be configured to control each air conditioner 130 via a communication path (infrared rays and other wireless means, for example) different from the network 150.
The learning model generation unit 500 generates or updates a learning model stored in the learning model storage unit 300 (machine learning), based on conditions, which are specified by the condition specification unit 110, in an environment managed by the air conditioning control system 1 and a feature amount which is created by the feature amount creation unit 210 and represents a feature of the environment managed by the air conditioning control system 1. This learning model generation unit 500 selects a learning model which is to be an object of generation or updating based on conditions, which are specified by the condition specification unit 110, in an environment managed by the air conditioning control system 1 and performs machine learning with respect to the selected learning model, based on a feature amount which is created by the feature amount creation unit 210 and represents a feature of the environment managed by the air conditioning control system 1. Examples of timing at which the learning model generation unit 500 performs learning include a timing at which an operator manually changes setting of each air conditioner 130. In this case, the learning model generation unit 500 performs generation or updating (machine learning) of a learning model by using a feature amount, which is created by the feature amount creation unit 210 and represents a feature of the environment managed by the air conditioning control system 1, as a state variable and using a set temperature and a wind direction, for example, of each air conditioner 130 as label data, with respect to a learning model selected based on conditions in the environment managed by the air conditioning control system 1.
In the case where a learning model associated with (a combination of) conditions, which are specified by the condition specification unit 110, in an environment managed by the air conditioning control system 1 is not stored in the learning model storage unit 300, the learning model generation unit 500 newly generates a learning model which is associated with (the combination) of the conditions. On the other hand, in the case where a learning model associated with (a combination of) conditions, which are specified by the condition specification unit 110, in an environment managed by the air conditioning control system 1 is stored in the learning model storage unit 300, the learning model generation unit 500 performs machine learning with respect to this learning model so as to update this learning model. In the case where a plurality of learning models associated with (combinations) of conditions, which are specified by the condition specification unit 110, in an environment managed by the air conditioning control system 1 are stored in the learning model storage unit 300, the learning model generation unit 500 may perform machine learning with respect to each of the learning models or may perform machine learning with respect to only a part of the learning models based on learning processing which can be executed by the learning model generation unit 500 and throughput of the learning model generation unit 500.
The learning model generation unit 500 may alter a learning model stored in the learning model storage unit 300 and generate a new learning model. As an example of the alteration of a learning model by the learning model generation unit 500, generation of a distilled model is exemplified. A distilled model is a learned model which is obtained such that learning is performed from the beginning in a machine learning device by using an output obtained with respect to an input into another machine learning device in which a learned model is incorporated. The learning model generation unit 500 can store a distilled model, which is obtained through such processing (called distillation processing), in the learning model storage unit 300 as a new learning model and can use the new learning model. A distilled model is generally smaller in size than an original learned model but exhibits the same accuracy as that of the original learned model, being more suitable for distribution to other computers through a network or the like. Another example of the alteration of a learning model by the learning model generation unit 500 is integration of learning models. In the case where structures of two or more learning models which are stored in a manner to be associated with (combinations of) conditions in an environment managed by the air conditioning control system 1 are similar to each other, for example, in the case where values of respective weight parameters are within a predetermined threshold value, the learning model generation unit 500 may integrate (the combinations of) conditions, which are associated with these learning models, in the environment managed by the air conditioning control system 1 and may store any of the two or more learning models, whose structures are similar to each other, in a manner to associate any of the two or more learning models with the integrated condition.
FIG. 3 is a functional block diagram schematically illustrating an air conditioning control system 1 according to a second embodiment.
In the air conditioning control system 1 according to the present embodiment, each functional block is mounted on one air conditioning controller 2 (which is to be structured on a central management device of air conditioners, a numerical controller, or the like). In such configuration, the air conditioning control system 1 according to the present embodiment infers a method for controlling the air conditioners 130 in an environment managed by the air conditioning control system 1 by using a learning model differing depending on installation states and operation states of the machines 120 in the environment managed by the air conditioning control system 1 so as to control the air conditioners 130 in the environment managed by the air conditioning control system 1. Further, one air conditioning controller 2 is capable of generating/updating learning models each corresponding to conditions in an environment managed by the air conditioning control system 1.
FIG. 4 is a functional block diagram schematically illustrating an air conditioning control system 1 according to a third embodiment.
In the air conditioning control system 1 according to the present embodiment, the environment management unit 100, the inference processing unit 200, and the air conditioning control unit 400 are mounted on the air conditioning controller 2, and the learning model storage unit 300 and the learning model generation unit 500 are mounted on a machine learning device 3 which is connected with the air conditioning controller 2 via standard interface and network. The machine learning device 3 may be mounted on a cell computer, a host computer, a cloud server, or a database server. In such configuration, inference processing using a learned model, which is relatively light processing, can be executed on the air conditioning controller 2 and processing for generating/updating a learning model, which is relatively heavy processing, can be executed on the machine learning device 3, so that the air conditioning control system 1 can be operated without interrupting an intrinsic action of the air conditioning controller 2.
FIG. 5 is a functional block diagram schematically illustrating an air conditioning control system 1 according to a fourth embodiment.
In the air conditioning control system 1 according to the present embodiment, the environment management unit 100 is mounted on the air conditioning controller 2, and the inference calculation unit 220, the learning model storage unit 300, and the learning model generation unit 500 are mounted on the machine learning device 3 which is connected with the air conditioning controller 2 via standard interface and network. Further, the air conditioning control unit 400 is separately prepared. In the air conditioning control system 1 according to the present embodiment, the configuration of the feature amount creation unit 210 is omitted on the assumption that a state amount detected by the state amount detection unit 140 is data which can be directly used for inference processing by the inference calculation unit 220 and generation/updating processing of a learning model performed by the learning model generation unit 500. In such configuration, inference processing using a learned model and generation/updating processing of a learning model can be executed on the machine learning device 3, so that the air conditioning control system 1 can be operated without interrupting an intrinsic action of the air conditioning controller 2.
FIG. 6 is a functional block diagram schematically illustrating an air conditioning control system 1 according to a fifth embodiment.
In the air conditioning control system 1 according to the present embodiment, each functional block is mounted on one air conditioning controller 2. In the air conditioning control system 1 according to the present embodiment, the configuration of the learning model generation unit 500 is omitted on the assumption that a plurality of learning models that have been learned and are associated with combinations of conditions in an environment managed by the air conditioning control system 1 are already stored in the learning model storage unit 300 and generation/updating of a learning model is not performed. In such configuration, the air conditioning control system 1 according to the present embodiment infers a method for controlling the air conditioners 130 in an environment managed by the air conditioning control system 1 by using a learning model differing depending on installation states and operation states of the machines 120 in the environment managed by the air conditioning control system 1, for example, so as to control the air conditioners 130 in the environment managed by the air conditioning control system 1. Further, since a learning model is not arbitrarily updated, this configuration is employable as the configuration of the air conditioning controller 2 which is shipped to customers, for example.
FIG. 7 is a functional block diagram schematically illustrating a modification of the air conditioning control system 1 according to the fifth embodiment.
The air conditioning control system 1 according to the present modification is an example in which the learning model storage unit 300 is mounted on an external storage 4 which is connected with the air conditioning controller 2 in the fifth embodiment (FIG. 6). In the present modification, learning models having larger capacity are stored in the external storage 4, so that more learning models can be used and learning models can be read without any intervention of a network or the like. Thus, this modification is beneficial when a real time property is required for inference processing.
FIG. 8 is a functional block diagram schematically illustrating an air conditioning control system 1 according to a sixth embodiment.
In the air conditioning control system 1 according to the present embodiment, the environment management unit 100 is mounted on the air conditioning controller 2, and the inference calculation unit 220 and the learning model storage unit 300 are mounted on the machine learning device 3 which is connected with the air conditioning controller 2 via standard interface and network. The machine learning device 3 may be mounted on a cell computer, a host computer, a cloud server, or a database server. In the air conditioning control system 1 according to the present embodiment, the configuration of the learning model generation unit 500 is omitted on the assumption that a plurality of learning models that have been learned and are associated with combinations of conditions in an environment managed by the air conditioning control system 1 are already stored in the learning model storage unit 300 and generation/updating of a learning model is not performed. In the air conditioning control system 1 according to the present embodiment, the configuration of the feature amount creation unit 210 is omitted on the assumption that a state amount detected by the state amount detection unit 140 is data which can be directly used for inference processing by the inference calculation unit 220. In such configuration, the air conditioning control system 1 according to the present embodiment infers a method for controlling the air conditioners 130 in an environment managed by the air conditioning control system 1 by using a learning model differing depending on installation states and operation states of the machines 120 in the environment managed by the air conditioning control system 1, for example, so as to control the air conditioners 130 in the environment managed by the air conditioning control system 1. Further, since a learning model is not arbitrarily updated, this configuration is employable as the configuration of the air conditioning controller 2 which is shipped to customers, for example.
FIG. 9 is a schematic flowchart of processing executed in the air conditioning control system 1 according to the present invention.
The flowchart illustrated in FIG. 9 shows an example of a flow of processing in the case where updating of a learning model is not performed in the air conditioning control system 1 (the fifth and sixth embodiments).
[Step SA01] The condition specification unit 110 specifies conditions in an environment managed by the air conditioning control system 1.
[Step SA02] The state amount detection unit 140 detects a state of the environment managed by the air conditioning control system 1 as a state amount.
[Step SA03] The feature amount creation unit 210 creates a feature amount representing a feature of the environment managed by the air conditioning control system 1 based on the state amount detected in step SA02.
[Step SA04] The inference calculation unit 220 selects and reads a learning model corresponding to the conditions, which are specified in step SA01, in the environment managed by the air conditioning control system 1 from the learning model storage unit 300 as a learning model to be used for inference.
[Step SA05] The inference calculation unit 220 infers a method for controlling the air conditioners 130 in the environment managed by the air conditioning control system 1 based on the learning model read in step SA04 and the feature amount created in step SA03.
[Step SA06] The air conditioning control unit 400 controls air conditioning based on the method for controlling air conditioning, which is inferred in step SA05.
FIG. 10 is a schematic flowchart of processing executed in the air conditioning control system 1 according to the present invention.
The flowchart illustrated in FIG. 10 shows an example of ies a flow of processing in the case where generation/updating of a learning model is performed in the air conditioning control system 1 (the first to fourth embodiments).
[Step SB01] The condition specification unit 110 specifies conditions in an environment managed by the air conditioning control system 1.
[Step SB02] The state amount detection unit 140 detects a state of the environment managed by the air conditioning control system 1 as a state amount.
[Step SB03] The feature amount creation unit 210 creates a feature amount representing a feature of the environment managed by the air conditioning control system 1 based on the state amount detected in step SB02.
[Step SB04] The inference calculation unit 220 selects and reads a learning model corresponding to the conditions, which are specified in step SB01, in the environment managed by the air conditioning control system 1 from the learning model storage unit 300 as a learning model to be used for inference.
[Step SB05] The learning model generation unit 500 determines whether or not a learning model that has been learned and corresponds to the conditions, which are specified in step SB01, in the environment managed by the air conditioning control system 1 is already generated in the learning model storage unit 300. If a learning model that has been learned is already generated, the processing proceeds to step SB07. If a learning model that has been learned is not generated yet, the processing proceeds to step SB06.
[Step SB06] The learning model generation unit 500 generates/updates a learning model corresponding to the conditions, which are specified in step SB01, in the environment managed by the air conditioning control system 1 based on the feature amount created in step SB03 and the processing proceeds to step SB01.
[Step SB07] The inference calculation unit 220 infers a method for controlling the air conditioners 130 in the environment managed by the air conditioning control system 1 based on the learning model read in step SB04 and the feature amount created in step SB03.
[Step SB08] The air conditioning control unit 400 controls air conditioning based on the method for controlling air conditioning, which is inferred in step SB05.
The embodiments of the present invention have been described above. The present invention is not limited to the examples of the above-described embodiments but may be embodied in various aspects by adding arbitrary alterations.

Claims

1. An air conditioning control system that controls an air conditioner in an environment in which at least one machine is installed, the air conditioning control system comprising:

a condition specification unit that specifies a condition in the environment;

a state amount detection unit that detects a state amount representing a state of the environment;

an inference calculation unit that infers a control method for the air conditioner in the environment based on the state amount;

an air conditioning control unit that controls the air conditioner based on the control method that is inferred by the inference calculation unit;

a learning model generation unit that generates or updates a learning model through machine learning using the state amount; and

a learning model storage unit that stores one or more learning models generated by the learning model generation unit in a manner such that the one or more learning model are associated with a combination of conditions specified by the condition specification unit, wherein

the inference calculation unit calculates a control method for the air conditioner in the environment managed by the air conditioning control system, by selectively using one or more learning models among learning models stored in the learning model storage unit, based on the condition in the environment that is specified by the condition specification unit.

2. The air conditioning control system according to claim 1, further comprising:

a feature amount creation unit that creates a feature amount characterizing the environment based on the state amount detected by the state amount detection unit, wherein

the inference calculation unit infers a control method for the air conditioner in the environment based on the feature amount, and

the learning model generation unit generates or updates learning model through machine learning using the feature amount.

3. The air conditioning control system according to claim 1, wherein the learning model generation unit alters an existing learning model stored in the learning model storage unit so as to generate a new learning model.

4. The air conditioning control system according to claim 1, wherein the learning model storage unit encrypts and stores a learning model generated by the learning model generation unit, and decrypts the encrypted learning model when the learning model encrypted is read by the inference calculation unit.

5. An air conditioning control system that controls an air conditioner in an environment in which one or more machines are installed the air conditioning control system comprising:

a condition specification unit that specifies a condition in the environment;

a state amount detection unit that detects a state amount representing the environment;

an air conditioning control unit that controls the air conditioner based on the control method that is inferred by the inference calculation unit; and

a learning model storage unit that stores at least one learning model that is preliminarily associated with a combination of conditions in the environment, wherein

the inference calculation unit calculates a control method for the air conditioner in the environment by selectively using one or more learning models among the learning models stored in the learning model storage unit, based on the condition in the environment that is specified by the condition specification unit

6. The air conditioning control system according to claim 5, further comprising:

a feature amount creation unit that creates a feature amount characterizing the environment based on the state amount, wherein

the inference calculation unit infers, based on the feature amount, a control method for the air conditioner in the environment managed by the air conditioning control system.

7. An air conditioning controller comprising:

the condition specification unit according to claim 1; and

the state amount detection unit according to claim 1.

8. An air conditioning controller comprising:

the condition specification unit according to claim 5; and

the state amount detection unit according to claim 5

9. An air conditioning control method comprising:

a step for specifying a condition for controlling an air conditioner in an environment in which one or more machines are installed;

a step for detecting a state amount representing the environment;

a step for inferring a control method for the air conditioner in the environment based on the state amount;

a step for controlling the air conditioner based on the control method; and

a step for generating or updating a learning model through machine learning using the state amount, wherein

in the step for inferring, a learning model to be used based on the condition in the environment that is specified in the step for specifying is selected from the one or more learning models that are preliminarily associated with a combination of conditions in the environment, and a control method for the air conditioner in the environment is calculated using the learning model that is selected.

10. The air conditioning control method according to claim 9, further comprising:

a step for creating a feature amount characterizing the environment based on the state amount, wherein

in the step for inferring, a control method for the air conditioner in the environment is inferred based on the feature amount, and

in the step for generating or updating a learning model, a learning model is generated or updated through machine learning using the feature amount.

11. An air conditioning control method comprising:

a step for detecting a state amount representing the environment;

a step for inferring a control method for the air conditioner in the environment based on the state amount; and

a step for controlling the conditioner based on the control method, wherein

in the step for inferring, a learning model to be used based on the condition in the environment that is specified in the step for specifying is selected from one or more learning models that are preliminarily associated with a combination of conditions in the environment, and a control method for the air conditioner in the environment is calculated using the learning model that is selected.

12. The air conditioning control method according to claim 11, further comprising:

in the step for inferring, a control method for the air conditioner in the environment is inferred based on the feature amount.

13. A learning model set comprising:

a plurality of learning models each of which is associated with a combination of conditions for controlling an air conditioner in an environment in which one or more machines are installed, wherein

each of the plurality of learning models is a learning model generated or updated, in a condition in the environment, based on a state amount representing the environment, and

one learning model is selected from the plurality of learning models based on a condition set in an environment, and

the learning model that is selected is used for processing of inferring a control method for the air conditioner in the environment.