Air conditioning control system

Abstract

In a case where real-time is required for decision of control content of an air conditioning device, a method using reinforcement learning has a problem of taking time. An air conditioning control system (100) transmits control content to an air conditioning device (30) to adjust an environmental state in an object space (81). The air conditioning control system (100) has an acquisition unit (11) and a control content decision unit (13). The acquisition unit (11) acquires a target environmental state which is a target environmental state. The control content decision unit (13) has a learning model which takes the target environmental state as input and takes a decision control content as output, the decision control content being control content to be transmitted to the air conditioning device (30) for bringing the object space (81) close to the target environmental state. The learning model is obtained by learning a learning control content which is control content for the air conditioning device (30) and a learning environmental state which is an environmental state in the object space (81) as a learning data set.

Description

TECHNICAL FIELD

[0001] The present application relates to an air conditioning control system. BACKGROUND

[0002] There is a technology that determines a control content of an air conditioning device in such a manner that an object space of an air conditioning action performed by the air conditioning device becomes a user-desired environmental state. In Patent Literature 1 (Japanese Patent Application Laid-Open No. 2019-522163), a control content of an air conditioning device is determined by a learning system including a reinforcement learning algorithm and a measured value from a sensor of various places within an object space. SUMMARY

[0003] PROBLEMS TO BE SOLVED BY THE INVENTION

[0004] In reinforcement learning, a reward function is determined in accordance with a user-desired environmental state, and a value function is learned using the reward function, and a control content of an air conditioning device is determined. Therefore, in a case where determination of a control content of an air conditioning device requires real-time, there is a problem that a method using reinforcement learning takes more time.

[0005] MEANS FOR SOLVING THE PROBLEMS

[0006] The air conditioning control system of the first aspect transmits a control content to an air conditioning device, and adjusts an environmental state in an object space of an air conditioning action performed by the air conditioning device. The air conditioning control system has an acquisition unit and a control content determination unit. The acquisition unit acquires a target environmental state that is a target of an environmental state. The control content determination unit has a learning model. The learning model takes the target environmental state as an input. The learning model takes a determined control content as an output, the determined control content being a control content that should be transmitted to the air conditioning device for bringing the object space close to the target environmental state. The learning model is obtained by learning a learning-use control content that is a control content for the air conditioning device and a learning-use environmental state that is an environmental state in the object space as a learning-use data set.

[0007] In the air conditioning control system of the first aspect, the air conditioning control system determines, by the learning model, a determined control content that should be transmitted to the air conditioning device for bringing the object space close to the target environmental state. The learning model is a learned model. Therefore, the air conditioning control system can determine the determined control content for becoming the target environmental state more in real time.

[0008] In the air-conditioning control system of the second aspect, the target environment state is an environment state targeted for a part of the subject space. The determined control content is control content to be transmitted to the air-conditioning device for bringing the part of the subject space close to the target environment state. The learning environment state is an environment state in the part of the subject space. Thus, the air-conditioning control system can accurately bring a part of the subject space to the environment state targeted.

[0009] In the air-conditioning control system of the second aspect, the target environment state is an environment state targeted for a part of the subject space. The determined control content is control content to be transmitted to the air-conditioning device for bringing the part of the subject space close to the target environment state. The learning environment state is an environment state in the part of the subject space. Thus, the air-conditioning control system can accurately bring a part of the subject space to the environment state targeted.

[0010] In the air-conditioning control system of the third aspect, in the air-conditioning control system of the second aspect, the part of the subject space is a prescribed three-dimensional region in the subject space.

[0011] In the air-conditioning control system of the third aspect, the part of the subject space is a prescribed three-dimensional region in the subject space. Thus, the air-conditioning control system can accurately bring a prescribed three-dimensional region in the subject space to the environment state targeted.

[0012] In the air-conditioning control system of the fourth aspect, in the air-conditioning control system of the second aspect, the part of the subject space is a prescribed two-dimensional region in the subject space.

[0013] In the air-conditioning control system of the fourth aspect, the part of the subject space is a prescribed two-dimensional region in the subject space. Thus, the air-conditioning control system can accurately bring a prescribed two-dimensional region in the subject space to the environment state targeted.

[0014] In the air-conditioning control system of the fifth aspect, in the air-conditioning control system of any one of the first to fourth aspects, the learning environment state is an output result obtained by simulating the learning control content as input one or more times.

[0015] In the air-conditioning control system of the fifth aspect, the air-conditioning control system simulates the learning control content as input one or more times, and acquires the learning environment state. Thus, the air-conditioning control system can easily acquire the learning data set.

[0016] In the air-conditioning control system of the sixth aspect, in the air-conditioning control system of any one of the first to fifth aspects, the control content is an air-conditioning control parameter including a quantity related to at least one of temperature, humidity, air direction, air volume, and air speed.

[0017] In the air-conditioning control system of the 7th aspect, the air-conditioning control system according to any one of the 1st to 6th aspects, the environmental state is an environmental parameter at one or more places in the object space. The environmental parameter includes an amount related to at least one of temperature, humidity, wind direction, air volume, and wind speed.

[0018] In the air-conditioning control system of the 8th aspect, the air-conditioning control system according to any one of the 1st to 7th aspects, the target environmental state is decided based on a desired input related to the environmental state via the user interface.

[0019] In the air-conditioning control system of the 8th aspect, the user sets the target environmental state via the user interface. Therefore, the user can easily set the desired target environmental state.

[0020] In the air-conditioning control system of the 9th aspect, the air-conditioning control system according to any one of the 1st to 8th aspects, the learning data set further includes spatial layout information as information of an object in the object space.

[0021] In the air-conditioning control system of the 9th aspect, the learning data set further includes spatial layout information as information of an object in the object space. Therefore, the information for the learning model to learn is increased, and the accuracy of the control content decided by the output of the learning model is improved.

[0022] In the air-conditioning control system of the 10th aspect, the air-conditioning control system according to the 9th aspect, the spatial layout information includes information related to heat as information of an object in the object space.

[0023] In the air-conditioning control system of the 11th aspect, the air-conditioning control system according to any one of the 1st to 10th aspects, the learning data set further includes configuration information of an air blowing port of one or more air conditioning devices.

[0024] In the air-conditioning control system of the 11th aspect, the learning data set further includes configuration information of an air blowing port of one or more air conditioning devices. Therefore, the information for the learning model to learn is increased, and the accuracy of the control content decided by the output of the learning model is improved. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1a is a schematic diagram of an air-conditioning control system.

[0026] Figure 1b is a schematic diagram of an air-conditioning control system.

[0027] Figure 1c is a schematic diagram of an air-conditioning control system.

[0028] Figure 1d is a schematic diagram of an air-conditioning control system.

[0029] Figure 2 is a configuration diagram of an air conditioner control system.

[0030] Figure 3a is a configuration diagram of an environmental state display process.

[0031] Figure 3b is a configuration diagram of a control content change process.

[0032] Figure 3c is a configuration diagram of a learning process.

[0033] Figure 3d is a configuration diagram of a learning process.

[0034] Figure 4 is a diagram showing a control content.

[0035] Figure 5 is a diagram showing a learning control content.

[0036] Figure 6 is a diagram after visualizing a temperature distribution in an object space using an environmental state (CFD).

[0037] Figure 7a is a diagram showing an environmental distribution of temperature.

[0038] Figure 7b is a diagram showing an environmental distribution of wind speed.

[0039] Figure 8a1 is a flowchart of an environmental state display process.

[0040] Figure 8a2 is a flowchart of an environmental state display process.

[0041] Figure 8b1 is a flowchart of a control content change process.

[0042] Figure 8b2 is a flowchart of a control content change process.

[0043] Figure 8c is a flowchart of a learning process.

[0044] Figure 9 is a summary diagram of a learning data set.

[0045] Figure 10 is a whole diagram of ResNet.

[0046] Figure 11 is a diagram of a step of imaging a three-dimensional temperature distribution. DETAILED DESCRIPTION

[0047] Hereinafter, the same environment state 41 is sometimes expressed with different representations, for example, as the environment state (AR) 41A, the environment state (CFD) 41C, and the environment state (environment profile) 41F. The environment state 41 refers to a state of temperature, humidity, and the like of a space that is an air conditioning target of the air conditioning device 30. The environment state (AR) 41A expresses the environment state 41 as visualization information for AR (Augmented Reality). The environment state (CFD) 41C simulates the environment state 41 and expresses it as CFD (Computational Fluid Dynamics) data. The environment state (environment profile) 41F expresses the environment state 41 with image data called an environment profile. Details thereof will be described later.

[0048] Further, with respect to the environment state 41, the former is expressed as a higher-order concept of the latter by attaching "target" or the like to the front of the latter as the target environment state 41T.

[0049] (1) Overall Configuration

[0050] The air conditioning control system 100 transmits a control content 42 to the air conditioning device 30, and adjusts the environment state 41 in an object space 81 in which the air conditioning device 30 performs an air conditioning operation. The control content 42 is an air conditioning control parameter that contains a quantity related to at least one of temperature, humidity, wind direction, wind volume, and wind speed. The object space 81 is, for example, an office in a building. The environment state 41 is an environmental parameter at one or a plurality of places in the object space 81. The environmental parameter contains a quantity related to at least one of temperature, humidity, wind direction, wind volume, and wind speed.

[0051] In the present embodiment, the air conditioning control system 100 adjusts the environment state 41 in a partial space 81a that is a part of the object space 81. The partial space 81a in the present embodiment is a prescribed two-dimensional region in the object space 81. In the present embodiment, the prescribed two-dimensional region is a plane of a prescribed height in the object space 81. However, it is not limited thereto, and the prescribed two-dimensional region is arbitrary.

[0052] As shown in FIG. 1, the air conditioning control system 100 mainly has an air conditioning control device 10. Figure 2

[0053] Figure 1a to Figure 1d is a schematic diagram of the air conditioning control system 100. As shown in FIG. 2, a user causes an AR application installed in a user terminal 20 to be started, and causes a camera of the user terminal 20 to face into an object space 81. Then, as shown in FIG. 3, an environment state 41 after AR visualization is displayed in a screen of the user terminal 20. In Figure 1a Figure 1b Figure 1b ​​​The image shows air blowing from the air conditioning unit 30. (Example) Figure 1c and Figure 1d As shown, a user can change the displayed wind direction to their desired direction by swiping their finger on the screen of the user terminal 20. The air conditioning control device 10 then determines the control content 42 of the air conditioning unit 30 to achieve the user's changed wind direction and sends this control content 42 to the air conditioning unit 30. As a result, the wind direction blowing from the air conditioning unit 30 changes to the direction changed by the user on the screen.

[0054] The air conditioning control system 100 in this embodiment performs environmental status display processing, control content change processing, and learning processing. Environmental status display processing is the process of displaying the environmental status 41 within the object space 81 on the screen of the user terminal 20 according to the control content 42 of the air conditioning device 30. Figure 3a The diagram shows the structure of the environmental status display process. The control content change process is the process of changing the control content 42 of the air conditioning device 30 in order to make the object space 81 the environmental state 41 desired by the user. Figure 3b The diagram shows the structure of the control content change process. The learning process is the process of generating a learning model 43 that determines the control content 42 of the air conditioning unit 30, which is used in the control content change process. Figure 3c A diagram showing the structure of the learning processes is provided. Details about these processes are explained in "(3) Processes".

[0055] (2) Detailed Structure

[0056] (2-1) Air conditioning control device

[0057] like Figure 2 As shown, the air conditioning control device 10 mainly includes an acquisition unit 11, a data conversion unit 12, a control content determination unit 13, an analog unit 14, a learning unit 15, and an AR content storage unit 16. Furthermore, the air conditioning control device 10 includes a control processing unit and a storage unit. The control processing unit can use a processor such as a CPU or a GPU. The control processing unit reads the program stored in the storage unit and performs prescribed image processing and calculation processing according to the program. Furthermore, the control processing unit can write the calculation results to the storage unit or read the information stored in the storage unit according to the program. Figure 2 The acquisition unit 11, data conversion unit 12, control content determination unit 13, simulation unit 14, learning unit 15, and AR content storage unit 16 shown are various functional blocks implemented by the control and arithmetic device.

[0058] The air conditioning control device 10 is connected to the user terminal 20 and the air conditioning unit 30 via a communication network 80 such as the Internet. In this embodiment, it is assumed that the air conditioning control device 10 is located in the cloud. However, the air conditioning control device 10 may also be located within the object space 81, and there is no particular limitation on its location.

[0059] (2-1-1) Obtaining the Department

[0060] exist Figure 3a In the environmental status display processing shown, the acquisition unit 11 acquires control content 42 from the air conditioning device 30. Figure 4 An example of control content 42 is shown. Figure 4 In the device, "20°C" is stored in the "Temperature" field, and "50%" is stored in the "Humidity" field. Values ​​representing the magnitude of each of the "Airflow" and "Wind Speed" are stored in five levels. When multiple air conditioning units 30 are present in the object space 81, the acquisition unit 11 acquires the control data 42 for each unit.

[0061] In addition, Figure 3b In the control content change processing shown, the acquisition unit 11 acquires the target environmental state 41, i.e., the target environmental state 41T, from the user terminal 20. The target environmental state 41 refers to the environmental state 41 specified by the user on the screen of the user terminal 20 that the user desires. In this embodiment, the target environmental state 41T is the target environmental state 41 in a plane at a predetermined height within the object space 81. In this embodiment, the acquisition unit 11 acquires the target environmental state 41T as visual information for AR, i.e., the target environmental state (AR) 41TA.

[0062] In addition, Figure 3c In the learning process shown, the acquisition unit 11 acquires the learning control content 42L. The learning control content 42L is used as the target variable of the learning model 43. Figure 5 An example of learning control content 42L is shown. Learning control content 42L is data obtained by combining various values ​​of temperature, humidity, etc., that the air conditioning unit 30 can obtain as setpoints. When there are multiple air conditioning units 30 in the object space 81, the acquisition unit 11 acquires the learning control content 42L of multiple units. The acquisition unit 11 acquires the learning control content 42L from, for example, a CSV file, a database server, etc.

[0063] (2-1-2) Simulation Department

[0064] exist Figure 3aIn the environmental state display processing shown, the simulation unit 14 simulates the environmental state 41 within the object space 81 based on the control content 42 of the air conditioning device 30 acquired by the acquisition unit 11. The simulation unit 14 outputs the environmental state (CFD) 41C as the simulation result. The environmental state (CFD) 41C is the simulated data related to the environmental parameters of each coordinate point within the object space 81. Figure 6 An example is shown where the temperature distribution within object space 81 is visualized using environmental conditions (CFD) 41C. Figure 6 The temperature distribution is shown in a plane 1m above the ground within object space 81. Figure 6 In the diagram, the darker areas of color represent the areas with higher temperatures.

[0065] In addition, Figure 3c In the learning process shown, the simulation unit 14 simulates the environmental state 41 within the object space 81 based on the learning control content 42L acquired by the acquisition unit 11. For example... Figure 5 As shown, the learning control content 42L consists of multiple control contents 42. Therefore, the simulation unit 14 simulates the environmental state 41 within the object space 81 for each control content 42. The learning environmental state 41L is the output result obtained by performing one or more simulations with the learning control content 42L as input. In this embodiment, the learning environmental state 41L is the environmental state 41 in a plane at a predetermined height within the object space 81. The simulation unit 14 outputs the learning environmental state (CFD) 41LC as a simulation result. The learning environmental state (CFD) 41LC consists of multiple CFD data corresponding to each simulation.

[0066] The above simulations can be performed using existing general-purpose simulation software, for example.

[0067] (2-1-3) Data Conversion Department

[0068] exist Figure 3a In the environmental status display processing shown, the data conversion unit 12 converts the environmental status (CFD) 41C output by the analog unit 14 into visual information for AR, namely environmental status (AR) 41A.

[0069] In addition, Figure 3bIn the control content change process shown, the data conversion unit 12 uses the target environment state (AR) 41TA acquired by the acquisition unit 11 and the environment state (CFD) 41C output by the simulation unit 14 in the environment state display process to generate the target environment state (CFD) 41TC. The target environment state (AR) 41TA is the AR-based visual information reflecting the environment state 41 desired by the user. The environment state (CFD) 41C is the environment state 41 before the user performs an operation on the screen of the user terminal 20, which does not reflect the user's desired environment state 41. The data conversion unit 12 uses the target environment state (AR) 41TA to update the environment state (CFD) 41C, thereby generating the environment state 41 desired by the user, i.e., the target environment state (CFD) 41TC.

[0070] When the acquisition unit 11 acquires multiple target environment states (AR) 41TAs from multiple user terminals 20 within a relatively short time, such as within 5 seconds, the data conversion unit 12 can also determine one target environment state (AR) 41TA based on these target environment states (AR) 41TAs using a prescribed method.

[0071] For example, suppose one user specifies the temperature of location A within object space 81 as 20°C, and another user specifies the same location A as 22°C. In the event of competition among multiple target environmental states (AR) 41TA, one target environmental state (AR) 41TA can be determined by averaging these multiple ARs 41TA. In the example above, the target environmental state (AR) 41TA with a temperature of 21°C for location A was determined. Furthermore, suppose one user specifies the temperature of location A within object space 81 as 20°C, and another user specifies the temperature of location B, which is different from location A, as 22°C. In the event of no competition among multiple target environmental states (AR) 41TA, a target environmental state (AR) 41TA that faithfully reflects the desires of each user can be determined. In the example above, the target environmental state (AR) 41TA with a temperature of 20°C for location A and a temperature of 22°C for location B was determined.

[0072] The data conversion unit 12 further converts the target environmental state (CFD) 41TC into a target environmental state (environmental distribution map) 41TF. In this embodiment, the environmental distribution map is a graph obtained by visualizing the temperature distribution, etc., in a plane at a predetermined height within the object space 81. The environmental distribution map is generated according to each environmental parameter, such as temperature and humidity. For example, in the case of a temperature environmental distribution map, the data conversion unit 12 obtains the temperature distribution in a plane at a predetermined height within the object space 81 from the target environmental state (CFD) 41TC. The data conversion unit 12 normalizes the obtained temperature distribution in the plane and then visualizes it. Normalization is a process of converting the temperature corresponding to each pixel into a value from 0 to 1. For example, if the maximum temperature of the temperature distribution is 25°C and the minimum temperature is 20°C, when the value obtained by subtracting 20°C from the temperature of each pixel is divided by 5°C (obtained by subtracting 20°C from 25°C), the temperature value of each pixel becomes a value from 0 to 1. Visualization is a process of assigning each pixel a shade corresponding to the magnitude of the normalized value to visualize it. Figure 7a An example of the temperature distribution in a plane 1m above the ground within object space 81 is shown. Furthermore, Figure 7b An example of an environmental distribution map of wind speed generated in the same manner is shown. The data conversion unit 12 outputs multiple environmental distribution maps corresponding to each environmental parameter, generated based on the target environmental state (CFD) 41TC, as the target environmental state (environmental distribution map) 41TF. Hereinafter, multiple environmental distribution maps generated based on one CFD data will be represented as a set of environmental distribution maps.

[0073] In addition, Figure 3c In the learning process shown, the data conversion unit 12 converts the learning environment state (CFD) 41LC output by the simulation unit 14 into a learning environment state (environmental distribution map) 41LF. As described above, since the learning environment state (CFD) 41LC is composed of multiple CFD data, the learning environment state (environmental distribution map) 41LF is composed of multiple sets of environmental distribution maps. The learning environment state (environmental distribution map) 41LF is used as an explanatory variable for the learning model 43.

[0074] Data transformation can be performed using simulation software that generates CFD data, or using programming languages ​​such as Python and R.

[0075] (2-1-4) Control Content Decision Department

[0076] The control content determination unit 13 has a learning model 43, which takes the target environmental state 41T as input and outputs the control content 42 to be sent to the air conditioning device 30 to make the object space 81 approach the target environmental state 41T, i.e., the determination control content 42D. Specifically, in Figure 3b In the control content change process shown, the control content determination unit 13 uses the learning model 43 obtained from the learning unit 15 to calculate and determine the control content 42D based on the target environmental state (environmental distribution map) 41TF. In this embodiment, the determined control content 42D is the control content 42 that should be sent to the air conditioning device 30 to make a plane at a specified height within the object space 81 approach the target environmental state 41T.

[0077] For example, the computation of learning model 43 can be performed using the functions of a machine learning platform that generates learning model 43.

[0078] (2-1-5) Academic Department

[0079] The learning unit 15 generates a learning model 43. The learning model 43 is obtained by learning from the control content 42 of the air conditioning device 30 (i.e., the learning control content 42L) and the environmental state 41 in the object space 81 (i.e., the learning environmental state 41L) as a learning dataset 44L. Specifically, in Figure 3c In the learning process shown, the learning unit 15 obtains the learning control content 42L from the acquisition unit 11 and the learning environment state (environmental distribution map) 41LF from the data conversion unit 12. The learning unit 15 uses the learning environment state (environmental distribution map) 41LF as the explanatory variable and the learning control content 42L as the target variable to generate a learning model 43.

[0080] The learning steps are explained in detail. Here, we describe the case where there is one air conditioning unit 30 within the object space 81. However, there can also be multiple air conditioning units 30 within the object space 81. Assume that the air conditioning control parameters constituting the control content 42 of the air conditioning unit 30 are temperature, airflow, and airflow direction. The image size of the environmental distribution map is 16 (vertical pixels) × 8 (horizontal pixels). Figure 9 This shows a summary of the learning dataset 44L. Figure 9 Two pieces of learning data are shown in the figure. Figure 9 On the left is the learning environment state (environmental distribution map) 41LF, which serves as the explanatory variable for the learning model 43. Here, two sets of environmental distribution maps are depicted. Since there are three air conditioning control parameters constituting the control content 42, as explained in "(2-1-3) Data Conversion Section", one set of environmental distribution maps consists of three environmental distribution maps. Figure 9 The right side is the learning control content 42L, which serves as the target variable for learning model 43.

[0081] In this embodiment, a regression model based on a Convolutional Neural Network (CNN) is used as the learning model 43. Furthermore, ResNet (Residual Network) is used as the CNN. Figure 10 This shows an overall diagram of ResNet. Figure 10 The ResNet consists of an input layer IPL, an intermediate layer IML, and an output layer OPL.

[0082] like Figure 10 As shown, image data of the environment distribution map is input to the input layer IPL of ResNet. A three-dimensional arrangement of (vertical pixels of the environment distribution map) × (horizontal pixels of the environment distribution map) × (number of environment distribution maps constituting one set) can be input to the input layer IPL of ResNet. This is to input one set of environment distribution maps at a time. Here, a three-dimensional arrangement of 16 (vertical pixels) × 8 (horizontal pixels) × 3 (number of environment distribution maps) is input to the input layer IPL.

[0083] ResNet's intermediate layer IML1 is composed of Figure 10 The residual block shown is composed of multiple convolutional layers and shortcut connections. Figure 10 In the computation of the residual block, the activation function is applied to the value obtained by adding the input from the upper convolutional layer and the output from the lower convolutional layer. The activation function used is the ReLU (Rectified Linear Unit) function.

[0084] like Figure 10 As shown, the intermediate layer IML2 of ResNet is a fully connected layer (FC). The activation function used is the ReLU function.

[0085] The control content 42 of the air conditioning unit 30 is output to the output layer OPL of the ResNet. A vector of dimensionality (number of air conditioning control parameters constituting the control content 42) × (number of air conditioning units 30) is output to the output layer OPL of the CNN. Here, a three-dimensional vector is output to the output layer OPL as 3 (number of air conditioning control parameters) × 1 (number of units). Figure 10 In the output layer OPL, the predicted values ​​of temperature, air volume, and wind direction are output sequentially from top to bottom in the three nodes (circles).

[0086] ResNet loss functions can include, for example, mean squared error.

[0087] Optimizations of ResNet include using probabilistic gradient descent.

[0088] In this embodiment, a ResNet was generated as the learning model 43. However, the learning model 43 could also be a CNN with other structures, a typical neural network, etc. The learning model 43 could be generated, for example, using a cloud-based machine learning platform.

[0089] (2-1-6) AR Content Storage Department

[0090] AR content storage unit 16 stores AR content 16C. AR content 16C consists of elements such as images used in AR. For example, such as... Figure 1b to Figure 1d As shown, AR content 16C is an image of wind displayed on the screen of user terminal 20.

[0091] AR content 16C is stored using the storage device provided with the air conditioning control device 10.

[0092] (2-2) User terminal

[0093] like Figure 2 As shown, the user terminal 20 mainly includes an AR processing unit 21, an input unit 22, and an output unit 23. Furthermore, the user terminal 20 includes a control and processing unit and a storage unit. The control and processing unit can use a processor such as a CPU or a GPU. The control and processing unit reads the program stored in the storage unit and performs prescribed image processing and calculations according to the program. Furthermore, the control and processing unit can write the calculation results to the storage unit or read the information stored in the storage unit according to the program. Figure 2 The AR processing unit 21, input unit 22, and output unit 23 shown are various functional blocks implemented by the control and arithmetic unit. In this embodiment, the user terminal 20 is assumed to be a tablet terminal or a smartphone terminal. However, the user terminal 20 may also be a PC, etc., and is not particularly limited. The user terminal 20 is connected to the air conditioning control device 10 and the air conditioning device 30 via a communication network 80 such as the Internet. The user terminal 20 is located within the object space 81.

[0094] (2-2-1) AR Processing Department

[0095] exist Figure 3a In the environmental status display processing shown, the AR processing unit 21 obtains environmental status (AR data) 41A and AR content 16C from the air conditioning control device 10. Furthermore, the AR processing unit 21 obtains image information 25 within the object space 81 from the input unit 22. The image information 25 is information obtained by the input unit 22 from the camera on the user terminal 20. The AR processing unit 21 combines these information to generate AR image information 24.

[0096] In addition, Figure 3bIn the control content change process shown, the AR processing unit 21 obtains AR image information 24 reflecting the environmental state 41 desired by the user from the input unit 22. The AR processing unit 21 extracts AR visualization information, i.e., target environmental state (AR) 41TA, from the AR image information 24.

[0097] The AR processing unit 21 is implemented, for example, through the functionality of an existing general-purpose AR application. The AR application is installed and used in the user terminal 20.

[0098] (2-2-2) Input Section

[0099] exist Figure 3a In the environmental status display processing shown, the input unit 22 obtains image information 25 within the object space 81 from the camera provided by the user terminal 20.

[0100] In addition, Figure 3b In the control content change process shown, the input unit 22 obtains AR image information 24 reflecting the user's desired environmental state 41 from the screen of the user terminal 20. The user specifies their desired environmental state 41 by swiping their finger across the AR image displayed on the screen of the user terminal 20. Figure 1c and Figure 1d This illustrates a scenario where a user interacts with an AR image representing wind direction using their finger. The AR image information 24 acquired by the input unit 22 is processed by the AR processing unit 21 into a target environmental state (AR) 41T. Therefore, the target environmental state 41T is determined based on desired input related to the environmental state 41 via the screen of the user terminal 20, which serves as the user interface.

[0101] (2-2-3) Output section

[0102] exist Figure 3a In the environmental status display processing shown, the output unit 23 obtains AR image information 24 from the AR processing unit 21 and outputs it to the screen of the user terminal 20. Thus, the user can view the environmental status 41 within the AR-visualized object space 81 through the screen of the user terminal 20. Figure 1b This illustrates a scenario where the environmental state 41 within the object space 81, visualized via AR, is displayed on the screen of the user terminal 20.

[0103] (2-3) Air conditioning unit

[0104] like Figure 2 As shown, the air conditioning unit 30 mainly includes a control unit 31. The air conditioning unit 30 is connected to the air conditioning control unit 10 and the user terminal 20 via a communication network 80 such as the Internet. The air conditioning unit 30 is located within the object space 81.

[0105] (2-3-1) Control Unit

[0106] The control unit 31 controls the temperature, humidity, and other properties of the air discharged from the air conditioning unit 30 according to the control content 42.

[0107] exist Figure 3a During the environmental status display processing shown, the control unit 31 sends control content 42 to the air conditioning control device 10.

[0108] In addition, Figure 3b In the control content change process shown, the control unit 31 obtains the determined control content 42D from the air conditioning control device 10. The control unit 31 controls the air conditioning device 30 according to the determined control content 42D, thereby making the target space 81 the environment state 41 desired by the user.

[0109] (3) Processing

[0110] As described above, the air conditioning control system 100 performs environmental status display processing, control content change processing, and learning processing. Each process will be explained in detail below.

[0111] (3-1) Environmental Status Display Processing

[0112] The environmental status display processing is the process of displaying the environmental status 41 within the object space 81 on the screen of the user terminal 20 based on the control content 42 of the air conditioning device 30. (Using...) Figure 8a1 and Figure 8a2 The flowchart illustrates the environmental status display process. Figure 8a1 and Figure 8a2 The flowchart is basically along Figure 3a The image shows an arrow.

[0113] When the AR application is launched as shown in step S1, the user terminal 20 requests the environmental status (AR) 41A and AR content 16C from the air conditioning control device 10 as shown in step S2.

[0114] When the air conditioning control device 10 receives a request from the user terminal 20, it obtains control content 42 from the air conditioning device 30 as shown in step S3.

[0115] When the air conditioning control device 10 obtains the control content 42 from the air conditioning device 30, as shown in step S4, it simulates the environmental state 41 in the object space 81 according to the control content 42 and generates an environmental state (CFD) 41C.

[0116] When the air conditioning control device 10 generates an environmental state (CFD) 41C, it converts the environmental state (CFD) 41C into an environmental state (AR) 41A as shown in step S5.

[0117] When the air conditioning control device 10 changes the environmental state (CFD) 41C to the environmental state (AR) 41A, it sends the environmental state (AR) 41A and AR content 16C to the user terminal 20 as shown in step S6.

[0118] When the user terminal 20 obtains the environmental status (AR) 41A and AR content 16C from the air conditioning control device 10, it acquires image information 25 from the camera of the user terminal 20 as shown in step S7.

[0119] When the user terminal 20 obtains the environmental state (AR) 41A, AR content 16C and image information 25, it generates AR image information 24 as shown in step S8.

[0120] When the user terminal 20 generates the AR image information 24, it displays the AR image information 24 on the screen of the user terminal 20 as shown in step S9.

[0121] (3-2) Handling of changes to control content

[0122] Control content change processing is the process of changing the control content 42 of the air conditioning device 30 in order to make the object space 81 the environmental state 41 desired by the user. Figure 8b1 and Figure 8b2 The flowchart illustrates the process of handling changes to control content. Figure 8b1 and Figure 8b2 The flowchart is basically along Figure 3b The diagram shows the arrows. Here, it is assumed that AR image information 24 is displayed on the screen of user terminal 20 through environmental status display processing.

[0123] When the user terminal 20 operates on the AR image on the screen as shown in step S1, it obtains AR image information 24 that reflects the environmental state 41 desired by the user as shown in step S2.

[0124] When the user terminal 20 obtains the AR image information 24, it extracts the target environment state (AR) 41TA from the AR image information 24 as shown in step S3.

[0125] When the user terminal 20 extracts the target environmental state (AR) 41TA, it sends the target environmental state (AR) 41TA to the air conditioning control device 10 as shown in step S4.

[0126] When the air conditioning control device 10 obtains the target environmental state (AR) 41TA, it uses the target environmental state (AR) 41TA and the environmental state (CFD) 41C generated in the environmental state display process to generate the target environmental state (CFD) 41TC as shown in step S5.

[0127] When the air conditioning control device 10 generates the target environmental state (CFD) 41TC, it converts the target environmental state (CFD) 41TC into the target environmental state (environmental distribution map) 41TF as shown in step S6.

[0128] When the air conditioning control device 10 obtains the target environmental state (environmental distribution map) 41TF, it uses the learning model 43 to calculate the control content 42D as shown in step S7.

[0129] When the air conditioning control device 10 calculates the decision control content 42D, it sends the decision control content 42D to the air conditioning device 30 as shown in step S8.

[0130] When the air conditioning device 30 obtains the decision control content 42D, it controls the air conditioning device 30 according to the decision control content 42D as shown in step S9.

[0131] When the air conditioning control device 10 calculates the control content 42D, it performs environmental status display processing based on the control content 42D as shown in step S10. Specifically, it will... Figure 8a1 Step S3 is changed to "obtaining the decision control content 42D", and environmental status display processing begins from step S3. The decision control content 42D is calculated based on the screen operated by the user, so the screen display is basically the same as the screen display after environmental status display processing based on the decision control content 42D. However, the decision control content 42D is predicted by the learning model 43, so the two screen displays are sometimes slightly different. In addition, the environmental status (CFD) 41C of the simulation unit 14 needs to be updated in advance to the environmental status corresponding to the decision control content 42D. Therefore, the air conditioning control device 10 performs environmental status display processing based on the decision control content 42D.

[0132] (3-3) Learning Processing

[0133] The learning process is the process of generating a learning model 43 that determines the control content 42 of the air conditioning unit 30, which is used in the control content change process. Figure 8c The flowchart illustrates the learning process. Figure 8c The flowchart is basically along Figure 3c The image shows an arrow.

[0134] As shown in step S1, the air conditioning control device 10 acquires the learning control content 42L.

[0135] When the air conditioning control device 10 obtains the learning control content 42L, it simulates the environmental state 41 in the object space 81 based on the learning control content 42L and generates the learning environment state (CFD) 41LC as shown in step S2.

[0136] When the air conditioning control device 10 generates the learning environment state (CFD) 41LC, it converts the learning environment state (CFD) 41LC into the learning environment state (environmental distribution map) 41LF as shown in step S3.

[0137] When the air conditioning control device 10 obtains the learning environment state (environmental distribution map) 41LF, it generates a learning model 43 by using the learning environment state (environmental distribution map) 41LF as an explanatory variable and the learning control content 42L as a target variable, as shown in step S4.

[0138] (4) Features

[0139] (4-1)

[0140] In order to make the object space 81 close to the environmental state 41 desired by the user, the existing air conditioning control system determines the control content 42 that should be sent to the air conditioning unit 30 through reinforcement learning.

[0141] However, in reinforcement learning, a reward function is determined based on the user's desired environmental state 41, and this reward function is used to learn a value function, thereby determining the control content 42 of the air conditioning device 30. Therefore, when the determination of the control content 42 of the air conditioning device 30 requires real-time processing, the use of reinforcement learning methods presents a relatively time-consuming issue.

[0142] The air conditioning control system 100 of this embodiment determines the control content 42 to be sent to the air conditioning unit 30 through the learned learning model 43. Therefore, the air conditioning control system 100 is faster than reinforcement learning and can determine the control content 42 to be sent to the air conditioning unit 30 in more real time.

[0143] (4-2)

[0144] In the air conditioning control system 100 of this embodiment, the target environmental state 41T is the target environmental state 41 in a portion of the target space 81, namely, the partial space 81a. The control content 42D is the control content 42 that should be sent to the air conditioning device 30 to bring the partial space 81a closer to the target environmental state 41T. The learning environmental state 41L is the environmental state 41 in the partial space 81a. Therefore, the air conditioning control system 100 can make a portion of the target space 81 accurately become the target environmental state 42.

[0145] (4-3)

[0146] In the air conditioning control system 100 of this embodiment, a portion of space 81a is a defined two-dimensional region within the target space 81. Therefore, the air conditioning control system 100 can make the defined two-dimensional region within the target space 81 a target environmental state 41.

[0147] (4-4)

[0148] The air conditioning control system 100 of this embodiment performs simulation to generate a learning dataset 44L for the learning model 43. Therefore, the air conditioning control system 100 can easily obtain the learning dataset 44L.

[0149] (4-5)

[0150] In the air conditioning control system 100 of this embodiment, the user sets the desired environmental state 41 by operating on the screen of the user terminal 20, which serves as the user interface. Therefore, the user can easily set the desired environmental state 41.

[0151] (5) Variations

[0152] (5-1) Variation 1A

[0153] The learning dataset 44L can also contain information about objects within the object space 81, i.e., spatial layout information 45. Spatial layout information 45 includes information related to the position or heat of objects as information about objects within the object space 81. The position of an object can be obtained, for example, from an object detection camera. Heat can be obtained, for example, from a thermal camera.

[0154] Furthermore, the learning dataset 44L may also include configuration information 46 for the air outlets of one or more air conditioning units 30. The configuration information 46 can be obtained when defining the object space 81.

[0155] The spatial layout information 45 and configuration information 46 are reflected in the learning environment status (environmental distribution map) 41LF. Specifically, in the simulation unit 14 for environmental status display processing and learning processing, the air conditioning control system 100 uses not only control content 42 but also spatial layout information 45 and configuration information 46 to generate CFD data. For example, as Figure 3d As shown, by using spatial layout information 45 and configuration information 46 in the simulation unit 14 of the learning process, the spatial layout information 45 and configuration information 46 are reflected in the learning environment state (environmental distribution map) 41LF.

[0156] By including spatial layout information 45 and configuration information 46 in the learning dataset 44L, the learning model 43 has more information to learn, and the air conditioning control system 100 can improve the accuracy of the decision control content 42D output by the learning model 43.

[0157] (5-2) Variation 1B

[0158] In this embodiment, such as Figure 3c As shown, the air conditioning control system 100 generates a learning environment state (environmental distribution map) 41LF through simulation. However, the learning environment state (environmental distribution map) 41LF can also be generated based on the actual measurement values ​​of various sensors corresponding to each environmental parameter, which are installed in the object space 81. For example, in order to generate a temperature environmental distribution map, the temperature distribution in a plane at a specified height above the ground is obtained based on the actual measurement values ​​of temperature sensors, etc. The method for generating a temperature environmental distribution map based on the temperature distribution in a plane at a specified height above the ground is as described in "(2-1-3) Data Conversion Unit".

[0159] Instead of relying on simulation, the learning environment state (environmental distribution map) 41LF is generated based on actual measured values ​​of environmental parameters. As a result, the air conditioning control system 100 can generate a more accurate learning environment state (environmental distribution map) 41LF. Consequently, the air conditioning control system 100 can improve the accuracy of the decision control content 42D output by the learning model 43.

[0160] (5-3) Variation 1C

[0161] The air conditioning control system 100 of this embodiment uses AR to allow the user to understand the current environmental state 41 within the object space 81 and specify the desired environmental state 41. However, to achieve the above purpose, the air conditioning control system 100 may also use VR (Virtual Reality), MR (Mixed Reality), and SR (Substitutional Reality).

[0162] (5-4) Variation 1D

[0163] In the air conditioning control system 100 of this embodiment, the partial space 81a is a defined two-dimensional region within the object space 81. However, the partial space 81a may also be a defined three-dimensional region within the object space 81. As a result, the air conditioning control system 100 can make the defined three-dimensional region within the object space 81 the target environmental state 41.

[0164] In this variation, the defined three-dimensional region is a three-dimensional rectangular region surrounding the workspace of a person, etc. However, it is not limited to this, and the defined three-dimensional region can be arbitrary.

[0165] (5-4-1) Obtaining the Department

[0166] In this modified example, the target environmental state 41T acquired by the acquisition unit 11 is the environmental state 41 in the three-dimensional rectangular region within the object space 81.

[0167] (5-4-2) Simulation Department

[0168] In this modified example, the learning environment state 41L simulated by the simulation unit 14 is the environment state 41 in the three-dimensional rectangular region within the object space 81.

[0169] (5-4-3) Data Conversion Department

[0170] In this modified example, the environmental distribution map generated by the data conversion unit 12 is a graphical representation of the temperature distribution and other parameters within a three-dimensional rectangular region of the object space 81. Figure 11 This is a diagram illustrating the steps involved in visualizing a three-dimensional temperature distribution. (For example...) Figure 11 As shown in the left figure, the data conversion unit 12 first divides the three-dimensional rectangular region into rectangular unit blocks. Figure 11 In the process, the three-dimensional rectangular region is divided into 5 (width) × 4 (height) × 4 (depth) unit blocks. Each unit block contains at least one simulated temperature point. Next, the data conversion unit 12 summarizes the temperature values ​​contained in the unit blocks. As a summary, for example, the average value of one or more temperatures contained in the unit block is calculated. The result is that one temperature is determined for each unit block. Then, as... Figure 11 As shown in the central diagram, the data conversion unit 12 divides the three-dimensional rectangular region along the height direction. Figure 11 In the central image, the three-dimensional rectangular region is divided into four blocks. Since each unit block corresponds to one temperature, therefore, as... Figure 11 As shown in the right image, each unit block can be viewed as four planes with temperature values ​​in each pixel. These are environmental distribution maps obtained by imagerizing the temperature distribution within a three-dimensional rectangular region. The environmental distribution maps are generated for each environmental parameter, such as temperature and humidity. Therefore, in Figure 11 For example, in generating environmental distribution maps related to temperature and humidity, since two environmental distribution maps (temperature and humidity) are generated for one of the aforementioned planes, a total of eight environmental distribution maps are generated. Generally, in a three-dimensional rectangular region of W (width) × H (height) × D (depth), when generating environmental distribution maps related to N environmental parameters, the number of environmental distribution maps in one set is H × N, and each environmental distribution map has W × D pixels. Furthermore, in the above example, the three-dimensional rectangular region is segmented in the height direction, but segmentation in the width or depth direction is also possible.

[0171] (5-4-4) Control Content Decision Department

[0172] In this modified example, the control content determination unit 13 calculates the control content 42D, which is the control content 42 to be sent to the air conditioning device 30 to make the three-dimensional rectangular area in the object space 81 approach the target environmental state 41T.

[0173] (5-4-5) Academic Department

[0174] In this variant, similar to the present embodiment, the learning unit 15 generates the learning model 43.

[0175] (5-5)

[0176] The embodiments of this disclosure have been described above, but it should be understood that various changes can be made to the methods and details without departing from the spirit and scope of this disclosure as set forth in the claims.

[0177] Label Explanation

[0178] 11: Acquisition Unit; 13: Control Content Determination Unit; 30: Air Conditioning Unit; 41: Environmental Status; 41L: Learning Environmental Status; 41T: Target Environmental Status; 42: Control Content; 42D: Determine Control Content; 42L: Learning Control Content; 43: Learning Model; 44L: Learning Dataset; 45: Spatial Layout Information; 46: Configuration Information; 81: Object Space; 81a: Partial Space; 100: Air Conditioning Control System.

[0179] Existing technical documents

[0180] Patent documents

[0181] Patent Document 1: Japanese Patent Publication No. 2019-522163

Claims

1. An air conditioning control system (100) that sends control content (42) to an air conditioning device (30) to adjust the environmental state (41) in the object space (81) where the air conditioning device performs air conditioning operations, wherein, The air conditioning control system (100) has the following features: The acquisition unit (11) acquires the target environmental state (41T), which is the environmental state as a target, and the target environmental state is represented by image data generated based on the current control content of the air conditioning device to simulate the environmental state in the object space. as well as The control content determination unit (13) uses a learning model (43) to generate control content (42D) based on the image data. The learning model (43) takes the image data as input and the control content (42D) as output. The control content (42D) is the control content to be sent to the air conditioning device to make the object space approach the target environmental state. The learning model is obtained by learning the control content (42L) and the environmental state (41L) as a learning dataset (44L). The control content (42L) is the control content for the air conditioning device, and the environmental state (41L) is generated based on the control content to simulate the environmental state in the object space. The environmental state is represented by the image data.

2. The air conditioning control system according to claim 1, wherein, The target environment state is the environment state that serves as the target within a partial space (81a), and the partial space (81a) is a part of the object space. The control content to be determined is the control content that should be sent to the air conditioning device to bring the partial space closer to the target environmental state. The learning environment state refers to the environmental state within the specified portion of the space.

3. The air conditioning control system according to claim 2, wherein, The partial space is a defined three-dimensional region within the object space.

4. The air conditioning control system according to claim 2, wherein, The partial space is a defined two-dimensional region within the object space.

5. The air conditioning control system according to any one of claims 1 to 4, wherein, The learning environment state is generated by performing multiple simulations using the learning control content.

6. The air conditioning control system according to any one of claims 1 to 4, wherein, The control content to be determined includes air conditioning control parameters that are related to at least one of temperature, humidity, wind direction, air volume, and wind speed.

7. The air conditioning control system according to any one of claims 1 to 4, wherein, The environmental state refers to environmental parameters at one or more locations within the object space. The environmental parameters include quantities related to at least one of temperature, humidity, wind direction, air volume, and wind speed.

8. The air conditioning control system according to any one of claims 1 to 4, wherein, The target environment state is determined based on the simulation and the desired input related to the environment state via the user interface.

9. The air conditioning control system according to any one of claims 1 to 4, wherein, The learning dataset also includes spatial layout information (45) that serves as information about objects within the object space.

10. The air conditioning control system according to claim 9, wherein, The spatial layout information includes heat-related information as information about the objects within the object space.

11. The air conditioning control system according to any one of claims 1 to 4, wherein, The learning dataset also includes configuration information (46) of one or more air outlets of the air conditioning device.

Images