JP6915371B2 - Control device, control program, learning data creation method, and learning method - Google Patents

Description

The present invention relates to a control device, a control program, a learning data creation method, and a learning method.

In recent years, machine learning such as deep learning may be used to acquire a function of controlling the operation of a desired device. For example, Patent Document 1 proposes an intelligent housing system using a neural device. Specifically, in Patent Document 1, a function of adjusting and controlling the electronic device according to the user's customary operating state by using identity information, coordinate information, and control information for which the user has adjusted the electronic device in the past is provided. A system for learning machine learning has been proposed.

Special Table 2016-532355

As described above, by using machine learning, it is possible to easily construct a system that has acquired the function of controlling the operation of a desired device. However, the present inventor has found that the following problems may occur in a system in which the function of controlling the operation of the controlled device is acquired by machine learning.

That is, when a plurality of different learners who have acquired the function of controlling the operation of the controlled object device by machine learning are used, a conflict may occur between a plurality of controls for one or a plurality of controlled devices. .. For example, when the same controlled object device is controlled by using the learner constructed for each user, when a control command for executing a different operation is issued from the learner of a different user, the controlled object is controlled. Conflicts occur in the control of the device. Further, for example, when controlling a plurality of controlled object devices at the same time by using a learner constructed according to each controlled object device, a control command is issued to execute an operation that cannot be realized at the same time, such as moving to the same place. If this is done, a conflict will occur in the control of the plurality of controlled devices. The present inventor has found that when such a control conflict occurs, the controlled device may become inoperable in the conventional system.

The present invention, on the one hand, has been made in view of such circumstances, and an object of the present invention is to prevent the controlled device from becoming inoperable even if a conflict occurs between a plurality of controls. To provide technology.

The present invention employs the following configuration in order to solve the above-mentioned problems.

That is, the control device according to one aspect of the present invention is based on the control value output from the learned first learner that has learned to control the operation of the first control target device. To the control value output from the first control processing unit that controls the operation of the control target device 1 and the learned second learner that has learned to control the operation of the second control target device. Based on this, the second control processing unit that controls the operation of the second control target device, the control of the first control target device based on the control value output from the first learner, and the second control target device. When the control of the second control target device based on the control value output from the learning device of the above conflicts with the control of the second control target device, the control of the first control target device and the second control target device can be modified. It is provided with a conflict resolving unit for resolving the conflict.

In this configuration, the first control processing unit uses the first learner to control the operation of the first controlled device. Further, the second control processing unit controls the operation of the second controlled target device by using the second learning device. Then, when a conflict occurs between the control of the first control target device and the control of the second control target device, the conflict resolution unit determines the first control target device and the second control target device. The conflict is resolved by modifying the control. Therefore, according to the configuration, even if a conflict occurs between a plurality of controls, the controlled device can be prevented from becoming inoperable. The first control target device and the second control target device may be the same control target device or may be different control target devices from each other. Further, the control target device may include any kind of device that can be a control target, and may be, for example, an air conditioner, a robot device, or the like.

In the control device according to the one aspect, the conflict resolving unit has the control value of the first controlled device output from the first learning device and the second learning device output from the second learning device. When the control value of the control target device is input, learning is performed so as to output the control value of the first control target device and the control value of the second control target device modified so as to eliminate the conflict. The conflict may be resolved by using the learned third learner. According to this configuration, even when the control of the controlled target device may conflict in a complicated manner, the conflict that may occur in the controlled target device can be easily eliminated.

The control device according to the one aspect is a control value of the first control target device output from the first learner and a control value of the second control target device output from the second learner. Based on the above, a conflict type identification unit that specifies conflict type information indicating how the controls of the first control target device and the second control target device compete with each other may be further provided, and the conflict resolution unit may be further provided. May further input the identified competition type information into the third learner. According to this configuration, it is possible to adopt a suitable conflict resolution method according to the competition method (type).

In the control device according to the one aspect, the competition type identification unit is the control value of the first control target device output from the first learning device and the second learning device output from the second learning device. When the control value of the control target device of is input, the conflict type information is specified by using the learned fourth learner which has been trained to output the output value corresponding to the conflict type information. May be good. According to this configuration, even when the control of the controlled target device may conflict in a complicated manner, the type of conflict that may occur in the controlled target device can be easily specified.

In the control device according to the one aspect, the first, second, third, and fourth learners may each be configured by a neural network. According to this configuration, it is possible to easily realize a control device capable of resolving a conflict that may occur in the control of the controlled device.

In the control device according to the one aspect, the conflict resolving unit controls the first controlled device based on the control value output from the first learner, and outputs from the second learner. The conflict may be resolved by giving priority to either one of the controls of the second controlled target device based on the control value. According to this configuration, one of the control value output from the first learner and the control value output from the second learner is prioritized and the other is ignored to control the controlled device. It is possible to surely eliminate the conflict that occurs.

In the control device according to the one aspect, the first control target device and the second control target device may be the same control target device, and the conflict resolving unit outputs from the first learner. The conflict may be eliminated by averaging the control value to be output and the control value output from the second learner. According to this configuration, by adopting the average value of the control value output from the first learner and the control value output from the second learner, the conflict that occurs in the control of the controlled device can be eliminated. Can be done.

As another form of the control device according to each of the above forms, an information processing method that realizes each of the above configurations may be used, a program may be used, or a computer or other device that records such a program. , A storage medium that can be read by a machine or the like. Here, the recording medium that can be read by a computer or the like is a medium that stores information such as a program by electrical, magnetic, optical, mechanical, or chemical action.

For example, in the control program according to one aspect of the present invention, the computer that controls the operation of the first control target device and the second control target device learns to control the operation of the first control target device. The step of acquiring the control value for controlling the first controlled target device output from the learned first learned device and the learning for controlling the operation of the second controlled target device are performed. The step of acquiring the control value for controlling the second controlled target device output from the second learned device that has been learned, and the first step based on the control value output from the first learner. When the control of the control target device 1 and the control of the second control target device based on the control value output from the second learner conflict with each other, the first is modified so as to eliminate the conflict. The step of acquiring the control value of the control target device 1 and the control value of the second control target device, and the first control target device and the second control target device based on the acquired control value. It is a program for executing the steps to be controlled.

Further, the learning data creation method according to one aspect of the present invention is the first control target output from the learned first learner that has learned to control the operation of the first control target device. The second control target device output from the learned second learner that has been trained to control the operation of the second control target device and the step of acquiring the control value for controlling the device. To the step of acquiring the control value for controlling, the control of the first control target device based on the control value output from the first learner, and the control value output from the second learner. The step of determining whether or not the control of the second control target device based on the above conflicts, the control of the first control target device based on the control value output from the first learner, and the second control of the first control target device. When the control of the second control target device based on the control value output from the learning device of the above conflicts with the control of the second control target device, the first control target device and the second control target device so as to eliminate the conflict. The step of determining the correction value of the control value, the control value obtained from the first learning device and the control value obtained from the second learning device are used as input data, and the determined correction value is used as teacher data. As a step, it is provided with a step of creating learning data for learning the learner. According to this configuration, it is possible to collect learning data for constructing the third learning device used for resolving the conflict that may occur in the control of the controlled device.

In the learning data creation method according to the above aspect, the correction value may be determined by the input of the operator. When the controlled target device is a device used by a person, it is possible to create optimal learning data for constructing a third learning device used for resolving a conflict that may occur in the control of the controlled target device.

In the learning data creation method according to the above aspect, the modified value may be determined according to a predetermined rule. According to this configuration, learning data used for constructing the third learning device can be easily created.

Further, the learning method according to one aspect of the present invention includes a step of acquiring the learning data created by the learning data creating method according to any one of the above forms, and a step of learning the learning device using the acquired learning data. , Equipped with. According to this configuration, it is possible to construct the third learning device used for resolving the conflict that may occur in the control of the controlled device.

According to the present invention, it is possible to provide a technique for preventing a controlled device from becoming inoperable even if a conflict occurs between a plurality of controls.

FIG. 1 schematically illustrates an example of an application scene of the control device and the learning device according to the embodiment. FIG. 2 schematically illustrates an example of the hardware configuration of the control device according to the embodiment. FIG. 3 schematically illustrates an example of the hardware configuration of the data collection control device according to the embodiment. FIG. 4 schematically illustrates an example of the hardware configuration of the learning device according to the embodiment. FIG. 5 schematically illustrates an example of the functional configuration of the control device according to the embodiment. FIG. 6 schematically illustrates an example of the functional configuration of the data collection control device according to the embodiment. FIG. 7 schematically illustrates an example of the functional configuration of the learning device according to the embodiment. FIG. 8 illustrates an example of the processing procedure of the control device according to the embodiment. FIG. 9 illustrates an example of the processing procedure of the data collection control device according to the embodiment. FIG. 10 illustrates an example of the processing procedure of the learning device according to the embodiment. FIG. 11 schematically illustrates an example of the configuration of the control device according to the modified example. FIG. 12 schematically illustrates an example of the configuration of the control device according to the modified example. FIG. 13 schematically illustrates an example of the configuration of the control device according to the modified example.

Hereinafter, embodiments according to one aspect of the present invention (hereinafter, also referred to as “the present embodiment”) will be described with reference to the drawings. However, the embodiments described below are merely examples of the present invention in all respects. Needless to say, various improvements and modifications can be made without departing from the scope of the present invention. That is, in carrying out the present invention, a specific configuration according to the embodiment may be appropriately adopted. Although the data appearing in the present embodiment is described in natural language, more specifically, it is specified in a pseudo language, a command, a parameter, a machine language, etc. that can be recognized by a computer.

§1 Application example First, an example of a situation in which the present invention is applied will be described with reference to FIG. FIG. 1 schematically illustrates an example of application situations of the control device 1 and the learning device 3 according to the present embodiment.

As shown in FIG. 1, the control device 1 according to the present embodiment controls the operation of the air conditioning device 4 which is a controlled target device according to instructions from a plurality of users (users A and B in FIG. 1). It is an information processing device. The air conditioner 4 is, for example, a known air conditioner that regulates the temperature in a room, and corresponds to the “first control target device” and the “second control target device” of the present invention. That is, in the present embodiment, the first controlled object device and the second controlled object device are the same. However, the first control target device and the second control target device do not have to be limited to such an example, and may be different devices.

The control device 1 according to the present embodiment includes two learning devices for controlling the operation of the air conditioner 4. The first learner (the first neural network 5 described later) has learned in advance the control of the operation of the air conditioner 4 according to the preference of the user A. On the other hand, the second learner (second neural network 6 described later) has learned in advance the control of the operation of the air conditioner 4 according to the preference of the user B. The control device 1 controls the operation of the air conditioner 4 based on the control values output from each of the first learner and the second learner.

At this time, if the control value output from the first learner and the control value output from the second learner are different, there is a possibility that a conflict may occur in the control of the operation of the air conditioner 4. For example, in a situation where the room temperature is 24 degrees, the control value output from the first learner constitutes a command to set the room temperature to 26 degrees, and the control value output from the second learner constitutes the room temperature. When the command to set the temperature to 22 degrees is configured, a conflict occurs in the control of the operation of the air conditioner 4.

Therefore, the control device 1 according to the present embodiment controls the air conditioner 4 based on the control value output from the first learner and controls the air conditioner 4 based on the control value output from the second learner. When there is a conflict, the conflict is resolved by modifying the control of the air conditioner 4. Specifically, the control device 1 according to the present embodiment uses a third learner (third neural network 7 described later) to eliminate the competition for control of the air conditioner 4.

When the control values output from the first learner and the second learner are input to the third learner, the control value modified so as to eliminate the conflict (hereinafter, also referred to as "corrected control value"). It has been learned in advance to output). Therefore, the control device 1 inputs the control values obtained from the first learner and the second learner into the third learner to obtain the control values corrected so as not to cause a conflict. Can be done. The control device 1 controls the operation of the air conditioner 4 based on the corrected control value thus obtained.

On the other hand, the learning device 3 according to the present embodiment is an information processing device that performs machine learning of the third learning device. The learning device 3 according to the present embodiment uses the data collection control device 2 to collect learning data to be used for machine learning of the third learning device. Similar to the control device 1, the data collection control device 2 uses the first learner and the second learner to operate the air conditioner 4 so as to suit the tastes of each user (A, B). To control. However, the data collection control device 2 is different from the control device 1 in that it does not resolve the control conflict of the air conditioner 4 (does not use the third learner).

That is, in the control of the air conditioner 4 by the data collection control device 2, the above-mentioned conflict may occur. Therefore, in the data collection control device 2, does the control of the air conditioner 4 based on the control value obtained from the first learner compete with the control of the air conditioner 4 based on the control value obtained from the second learner? Judge whether or not. When it is determined that the control of the air conditioner 4 based on the control value obtained from the first learner and the control of the air conditioner 4 based on the control value obtained from the second learner conflict with each other, the data collection control device 2 , Determine the correction value of the control value so as to resolve the conflict.

For example, the data collection control device 2 gives priority to either one of the control values obtained from each of the first learner and the second learner. That is, the data collection control device 2 handles the priority control value as the corrected control value. Further, for example, the data collection control device 2 calculates the average value of the control values obtained from each of the first learner and the second learner as the corrected control value. As a result, the data collection control device 2 can acquire the corrected control value determined to eliminate the conflict.

Then, the data collection control device 2 uses the control values obtained from the first learner and the second learner as input data, and the corrected control value obtained as described above as the teacher data, as the third learner. Create learning data to be used for machine learning. That is, the data collection control device 2 creates learning data by combining each control value before modification and the modified control value.

The learning device 3 acquires the learning data created in this way, and performs machine learning of the third learning device using the acquired learning data, so that the learned third device that can be used in the control device 1 can be used. Build the learning device of 3. The control device 1 may acquire a third learned device that has been learned from the learning device 3 via a network, for example. Further, when the control device 1 is manufactured, the third learned device that has been learned as embedded data may be incorporated into the control device 1.

As described above, the control device 1 according to the present embodiment uses the learned first learner and the second learner, so that the air conditioner 4 is suitable for the preference of each user (A, B). You can control the operation of. In addition, when there is a conflict in the control of the air conditioner 4 based on the control values obtained from each of the first learner and the second learner, the third learner is used to eliminate the conflict. Can be done. Therefore, according to the present embodiment, it is possible to prevent the air conditioner 4 from becoming inoperable even if a conflict occurs between the controls by each user (A, B).

§2 Configuration example [Hardware configuration]
<Control device>
Next, an example of the hardware configuration of the control device 1 according to the present embodiment will be described with reference to FIG. FIG. 2 schematically illustrates an example of the hardware configuration of the control device 1 according to the present embodiment.

As shown in FIG. 2, the control device 1 according to the present embodiment is a computer to which the control unit 11, the storage unit 12, and the external interface 13 are electrically connected. In FIG. 2, the external interface is described as "external I / F".

The control unit 11 includes a CPU (Central Processing Unit), a RAM (Random Access Memory), a ROM (Read Only Memory), and the like, and is configured to execute various information processing based on programs and data. The storage unit 12 has a control program 121 executed by the control unit 11, a first motion control learning result data 122 showing information about the learned first learning device, and a second learning device showing information about the learned second learning device. 2 The motion control learning result data 123, the conflict elimination learning result data 124 indicating information about the learned third learning device, and the like are stored.

The control program 121 is a program for causing the control unit 11 to execute a process (FIG. 8) for controlling the operation of the air conditioner 4 described later. The first motion control learning result data 122 is data used for setting the learned first learner. The second motion control learning result data 123 is data used for setting the learned second learner. The conflict resolution learning result data 124 is data used for setting the learned third learner. Details will be described later.

The external interface 13 is an interface for connecting to an external device, and is appropriately configured according to the external device to be connected. In the present embodiment, the control device 1 is connected to the air conditioner 4 via the external interface 13. The control device 1 may be connected to a drive device or the like for reading data stored in the storage medium via the external interface 13. In this case, the control device 1 may acquire the control program 121, the first motion control learning result data 122, the second motion control learning result data 123, and the conflict resolution learning result data 124 via the drive device. .. When using the drive device, the control program 121, the first motion control learning result data 122, the second motion control learning result data 123, and the conflict resolution learning result data 124 may be stored in the storage medium. .. The storage medium stores the information of the program or the like by electrical, magnetic, optical, mechanical or chemical action so that the information of the program or the like recorded by the computer or other device or machine can be read. It is a medium. The storage medium is, for example, a CD (Compact Disk), a DVD (Digital Versatile Disk), a flash memory, or the like.

<Control device for data collection>
Next, an example of the hardware configuration of the data collection control device 2 according to the present embodiment will be described with reference to FIG. FIG. 3 schematically illustrates an example of the hardware configuration of the data collection control device 2 according to the present embodiment.

As shown in FIG. 3, the data collection control device 2 according to the present embodiment is a control device used in the scene of collecting learning data, and is configured in substantially the same manner as the control device 1. That is, the data collection control device 2 is a computer to which the control unit 21, the storage unit 22, and the external interface 23 are electrically connected. In FIG. 3, the external interface is described as "external I / F" as in FIG.

The control unit 21, the storage unit 22, and the external interface 23 are configured in the same manner as the control unit 11, the storage unit 12, and the external interface 13 of the control device 1. However, the storage unit 22 may execute the data collection control program 221, the first motion control learning result data 122, the second motion control learning result data 123, the data collection control program 221 to create learning data 223, etc. Remember.

The data collection control program 221 is a program for causing the data collection control device 2 to execute the learning data collection process (FIG. 9) described later. When the learning data 223 inputs the control values output from the first learning device and the second learning device, the learning data 223 learns the third learning device so as to output the corrected control values so as to eliminate the conflict. It is the data to do. Details will be described later.

<Learning device>
Next, an example of the hardware configuration of the learning device 3 according to the present embodiment will be described with reference to FIG. FIG. 4 schematically illustrates an example of the hardware configuration of the learning device 3 according to the present embodiment.

As shown in FIG. 4, the learning device 3 according to the present embodiment is a computer to which the control unit 31, the storage unit 32, the communication interface 33, the input device 34, the output device 35, and the drive 36 are electrically connected. .. In FIG. 4, the communication interface is described as "communication I / F".

The control unit 31 includes a CPU, RAM, ROM, and the like, and is configured to execute various information processing based on programs and data. The storage unit 32 stores the learning program 321 executed by the control unit 31, the learning data 223 used for learning of the third learning device, the conflict resolution learning result data 124 created by executing the learning program 321 and the like. .. The learning program 321 is a program for causing the learning device 3 to execute a learning process (FIG. 10) described later.

The communication interface 33 is, for example, a wired LAN (Local Area Network) module, a wireless LAN module, or the like, and is an interface for performing wired or wireless communication via a network. The input device 34 is, for example, a device for inputting a mouse, a keyboard, or the like. The output device 35 is, for example, a device for outputting a display, a speaker, or the like.

The drive 36 is, for example, a CD drive, a DVD drive, or the like, and is a drive device for reading a program stored in the storage medium 91. The type of the drive 36 may be appropriately selected according to the type of the storage medium 91. The learning program 321 and / or the learning data 223 may be stored in the storage medium 91.

The storage medium 91 stores the information of the program or the like by electrical, magnetic, optical, mechanical or chemical action so that the information of the program or the like recorded by the computer or other device, the machine or the like can be read. It is a medium to do. The learning device 3 may acquire the learning program 321 and / or the learning data 223 from the storage medium 91.

Note that FIG. 4 illustrates a disc-type storage medium such as a CD or DVD as an example of the storage medium 91. However, the type of the storage medium 91 is not limited to the disc type, and may be other than the disc type. Examples of storage media other than the disk type include semiconductor memories such as flash memories.

[Functional configuration]
<Control device>
Next, an example of the functional configuration of the control device 1 according to the present embodiment will be described with reference to FIG. FIG. 5 schematically illustrates an example of the functional configuration of the control device 1 according to the present embodiment.

The control unit 11 of the control device 1 expands the control program 121 stored in the storage unit 12 into the RAM. Then, the control unit 11 interprets and executes the control program 121 expanded in the RAM by the CPU to control each component. As a result, as shown in FIG. 5, the control device 1 according to the present embodiment functions as a computer including the first control processing unit 111, the second control processing unit 112, and the conflict resolving unit 113.

The first control processing unit 111 controls the operation of the air conditioner 4 by using the first neural network 5 which is the first learner. The first neural network 5 has learned in advance the control of the operation of the air conditioner 4 suitable for the preference of the user A. The first control processing unit 111 acquires the control value for the air conditioner 4 from the first neural network 5 by inputting the instruction data, the position information, and the like from the user A into the first neural network 5. ..

On the other hand, the second control processing unit 112 controls the operation of the air conditioner 4 by using the second neural network 6 which is the second learner. The second neural network 6 has learned in advance the control of the operation of the air conditioner 4 suitable for the preference of the user B. The second control processing unit 112 acquires the control value for the air conditioner 4 from the second neural network 6 by inputting the instruction data, the position information, and the like from the user B into the second neural network 6. ..

The type of information (data) to be input to the first neural network 5 and the second neural network 6 may be appropriately determined according to the embodiment. Users A and B may request the air conditioner 4 to adjust the room temperature by using, for example, a user terminal such as a PC (Personal Computer), a mobile phone, or a remote controller. In response to this, the control device 1 may receive instruction data from the user terminals of each user (A, B) by known wireless or wired data communication.

At this time, the control device 1 may acquire various information to be input to each neural network (5, 6) along with the instruction data from the user terminal. For example, when the user terminal includes a GPS (Global Positioning System) signal receiver, the control device 1 receives information to be input to each neural network (5, 6) from each user terminal to each user (A). , B) may be acquired.

Further, the control device 1 may store the personal information of each user (A, B) in the storage unit 12 in advance. In this case, when the control unit 11 receives the instruction data from each user terminal, the control unit 11 inputs the personal information of each user (A, B) from the storage unit 12 as the information to be input to each neural network (5, 6). You may get it.

When the conflict resolving unit 113 conflicts between the control of the air conditioner 4 based on the control value output from the first neural network 5 and the control of the air conditioner 4 based on the control value output from the second neural network 6. In addition, the conflict is resolved by modifying the control of the air conditioner 4. In the present embodiment, the third neural network 7, which is the third learning device, is used to resolve the conflict.

When the third neural network 7 inputs the control values output from each of the first neural network 5 and the second neural network 6, the third neural network 7 learns in advance to output the corrected control values so as to eliminate the conflict. It's done. Therefore, the conflict resolving unit 113 has been modified so as to resolve the conflict by inputting the control values output from each of the first neural network 5 and the second neural network 6 into the third neural network 7. The control value can be obtained from the third neural network 7.

Next, each neural network 5 to 7 will be described. As shown in FIG. 5, the first neural network 5 is a multi-layered neural network used for so-called deep learning, and the input layer 51, the intermediate layer (hidden layer) 52, and the output layer 53 are arranged in order from the input. I have.

In the example of FIG. 5, the first neural network 5 includes an intermediate layer 52 of one layer, the output of the input layer 51 is the input of the intermediate layer 52, and the output of the intermediate layer 52 is the output layer 53. It is an input. However, the number of the intermediate layers 52 is not limited to one layer, and the first neural network 5 may include two or more intermediate layers 52.

Each layer 51-53 comprises one or more neurons. For example, the number of neurons in the input layer 51 can be set according to the number of pieces of information used for input. The number of neurons in the middle layer 52 can be appropriately set according to the embodiment. Further, the number of neurons in the output layer 53 can be set according to the number of types of control values to be output.

Neurons in adjacent layers are appropriately connected to each other, and a weight (connection load) is set for each connection. In the example of FIG. 5, each neuron is connected to all neurons in the adjacent layer, but the connection of neurons does not have to be limited to such an example and is appropriately set according to the embodiment. You can.

A threshold is set for each neuron, and basically, the output of each neuron is determined by whether or not the sum of the products of each input and each weight exceeds the threshold. The first control processing unit 111 inputs various information such as instruction data and position information from the user A to the input layer 51 of the first neural network 5, and each layer 51 to 53 in the forward propagation direction. A control value (output value) can be obtained from the output layer 53 by determining the firing of each neuron included in.

The configuration of the first neural network 5 as described above (for example, the number of layers of the neural network, the number of neurons in each layer, the connection relationship between neurons, the transfer function of each neuron), the weight of the connection between each neuron, And the information indicating the threshold value of each neuron is included in the first motion control learning result data 122. The first control processing unit 111 refers to the first motion control learning result data 122 to set the first neural network 5 that has learned to control the motion of the air conditioner 4 that suits the preference of the user A. conduct.

The second neural network 6 and the third neural network 7 are configured in the same manner as the first neural network 5. That is, the second neural network 6 includes an input layer 61, an intermediate layer (hidden layer) 62, and an output layer 63 in this order from the input. The third neural network 7 includes an input layer 71, an intermediate layer (hidden layer) 72, and an output layer 73 in this order from the input. The number of each intermediate layer (62, 72), the number of neurons in each layer (61-63, 71-73), and the connection of neurons in adjacent layers may be appropriately set according to the embodiment.

The second control processing unit 112 inputs various information such as instruction data and position information from the user B to the input layer 61 of the second neural network 6, and is included in each of the layers 61 to 63 in the forward propagation direction. A control value (output value) can be obtained from the output layer 63 by determining the firing of each neuron. Further, the conflict resolving unit 113 inputs the control values output from the output layers (53, 63) of the first neural network 5 and the second neural network 6 to the input layer 71, and each layer in the forward propagation direction. By determining the firing of each neuron included in 71 to 73, a corrected control value (output value) can be obtained from the output layer 73.

Information indicating the configuration of the second neural network 6 as described above, the weight of the connection between each neuron, and the threshold value of each neuron is included in the second motion control learning result data 123. The second control processing unit 112 refers to the second motion control learning result data 123, and sets the second neural network 6 that has learned the control of the motion of the air conditioner 4 suitable for the preference of the user B. conduct.

Similarly, the information indicating the configuration of the third neural network 7 as described above, the weight of the connection between each neuron, and the threshold value of each neuron is included in the conflict resolution learning result data 124. The conflict resolution unit 113 has been modified so as to resolve the conflict when the control values output from each of the first neural network 5 and the second neural network 6 are input with reference to the conflict resolution learning result data 124. The third neural network 7 that has been learned is set so as to output the control value.

<Control device for data collection>
Next, an example of the functional configuration of the data collection control device 2 according to the present embodiment will be described with reference to FIG. FIG. 6 schematically illustrates an example of the functional configuration of the data collection control device 2 according to the present embodiment.

The control unit 21 of the data collection control device 2 expands the data collection control program 221 stored in the storage unit 22 into the RAM. Then, the control unit 21 interprets and executes the data collection control program 221 expanded in the RAM by the CPU to control each component. As a result, as shown in FIG. 6, the data collection control device 2 according to the present embodiment has the first control processing unit 211, the second control processing unit 212, the correction value determination unit 213, and the learning data creation unit. It functions as a computer equipped with 214.

The first control processing unit 211 is the same as the first control processing unit 111 of the control device 1. That is, the first control processing unit 211 sets the first neural network 5 with reference to the first motion control learning result data 122. Then, the first control processing unit 211 inputs various information such as instruction data and position information from the user A to the input layer 51 of the set first neural network 5, and each layer 51 to 51 in the forward propagation direction. By determining the firing of each neuron included in the 53, the control value (output value) for the air conditioner 4 according to the preference of the user A is acquired from the output layer 53.

The second control processing unit 212 is the same as the second control processing unit 112 of the control device 1. That is, the second control processing unit 212 sets the second neural network 6 with reference to the second motion control learning result data 123. Then, the second control processing unit 212 inputs various information such as instruction data and position information from the user B to the input layer 61 of the set second neural network 6, and each layer 61 to 61 in the forward propagation direction. By determining the firing of each neuron included in 63, the control value (output value) for the air conditioner 4 according to the preference of the user B is acquired from the output layer 63.

The data collection control device 2 controls the operation of the air conditioner 4 based on the control values obtained from each of the first neural network 5 and the second neural network 6. However, if there is a conflict in the control of the air conditioner 4 as a result of trying to control the operation of the air conditioner 4, the air conditioner 4 may become inoperable.

For example, as described above, in a situation where the room temperature is 24 degrees, the control value output from the first neural network 5 constitutes a command to raise the room temperature to 26 degrees, and is output from the second neural network 6. It is assumed that the control value constitutes a command to lower the room temperature to 22 degrees. In such a case, a conflict arises in the control of the air conditioner 4, and the data collection control device 2 should control the operation of the air conditioner 4 so as to raise the room temperature, or the air conditioner 4 so as to lower the room temperature. It becomes impossible to judge whether to control the operation of.

Therefore, when a conflict occurs in the control of the air conditioner 4 as a result of trying to control the operation of the air conditioner 4, the correction value determination unit 213 determines the correction value of the control value so as to eliminate the conflict. Then, the learning data creation unit 214 uses the control values obtained from each neural network (5, 6) as input data, and the corrected control value determined by the correction value determination unit 213 as teacher data, and uses the third neural network as the third neural network. The learning data 223 for constructing the network 7 is created.

<Learning device>
Next, an example of the functional configuration of the learning device 3 according to the present embodiment will be described with reference to FIG. 7. FIG. 7 schematically illustrates an example of the functional configuration of the learning device 3 according to the present embodiment.

The control unit 31 of the learning device 3 expands the learning program 321 stored in the storage unit 32 into the RAM. Then, the control unit 31 interprets and executes the learning program 321 expanded in the RAM by the CPU to control each component. As a result, as shown in FIG. 7, the learning device 3 according to the present embodiment functions as a computer including the learning data acquisition unit 311 and the learning processing unit 312.

The learning data acquisition unit 311 acquires the learning data 223 created above. The learning processing unit 312 constructs a third neural network used by the control device 1 by using the acquired learning data 223 and the neural network 8 for learning. That is, when the control value obtained from each neural network (5, 6) is input, the learning processing unit 312 trains the neural network 8 so as to output the corrected control value so as to eliminate the conflict.

The neural network 8 to be learned is configured in the same manner as the third neural network 7. That is, the neural network 8 for learning includes an input layer 81, an intermediate layer (hidden layer) 82, and an output layer 83, and each layer 81 to 83 is the same as each layer 71 to 73 of the third neural network 7. It is composed.

When the learning processing unit 312 inputs the control value obtained from each neural network (5, 6) to the input layer 81 by the learning process of the neural network, the learning processing unit 312 inputs the corrected control value from the output layer 83 so as to eliminate the conflict. A neural network 8 to be output is constructed. The neural network 8 constructed thereby can be used as a trained third neural network 7. The learning processing unit 312 stores information indicating the configuration of the constructed neural network 8, the weight of the connection between each neuron, and the threshold value of each neuron as the conflict resolution learning result data 124 in the storage unit 32.

<Others>
Each function of the control device 1, the data collection control device 2, and the learning device 3 will be described in detail in an operation example described later. In this embodiment, an example in which each function of the control device 1, the data collection control device 2, and the learning device 3 is realized by a general-purpose CPU has been described. However, some or all of the above functions may be realized by one or more dedicated hardware processors. Further, with respect to the functional configurations of the control device 1, the data collection control device 2, and the learning device 3, the functions may be omitted, replaced, or added as appropriate according to the embodiment.

§3 Operation example [Control device]
Next, an operation example of the control device 1 will be described with reference to FIG. FIG. 8 is a flowchart illustrating an example of the processing procedure of the control device 1. The processing procedure described below is only an example, and each processing may be changed as much as possible. Further, with respect to the processing procedure described below, steps can be omitted, replaced, and added as appropriate according to the embodiment.

(Start-up)
First, the system including the control device 1 and the air conditioner 4 is appropriately started. When the system is started, the control device 1 reads the control program 121 and executes the initial setting process. Specifically, the control unit 11 sets the structure of each neural network 5 to 7, the weight of the connection between each neuron, and the threshold value of each neuron with reference to the learning result data 122 to 124. Then, the control unit 11 controls the operation of the air conditioner 4 according to the following processing procedure.

(Step S101)
In step S101, the control unit 11 acquires various information used for controlling the operation of the air conditioner 4, in other words, information that is a factor for determining the operation of the air conditioner 4 from each user (A, B). As described above, the type of information used for controlling the operation of the air conditioner 4 may be appropriately determined according to the embodiment. Each user (A, B) may request the air conditioner 4 to adjust the room temperature by using, for example, a user terminal such as a PC (Personal Computer), a mobile phone, or a remote controller. In response to this, the control unit 11 may acquire various information such as instruction data and position information used for controlling the operation of the air conditioner 4 from the user terminal by known wireless or wired data communication. Further, the control unit 11 may acquire personal information of each user (A, B) from the storage unit 12 as information used for controlling the operation of the air conditioner 4.

(Step S102)
In the next step S102, the control unit 11 functions as the first control processing unit 111, and inputs various information such as instruction data and position information acquired from the user A to the first neural network 5. The first neural network 5 has been learned in advance so that when various information acquired from the user A is input, a control value for the air conditioner 4 according to the preference of the user A is output. Therefore, the control unit 11 inputs various information acquired from the user A to each neuron of the input layer 51, and determines the firing of each neuron included in each of the layers 51 to 53 in the direction of forward propagation. The control value for the air conditioner 4 according to the preference of A can be obtained from each neuron of the output layer 53.

Further, the control unit 11 functions as a second control processing unit 112, and inputs various information such as instruction data and position information acquired from the user B to the second neural network 6. The second neural network 6 has been learned in advance so that when various information acquired from the user B is input, the control value for the air conditioner 4 according to the preference of the user B is output. Therefore, the control unit 11 inputs various information acquired from the user B to each neuron of the input layer 61, and determines the firing of each neuron included in each of the layers 61 to 63 in the direction of forward propagation. The control value for the air conditioner 4 according to the preference of B can be obtained from each neuron of the output layer 63.

(Step S103)
In the next step S103, the control unit 11 functions as the conflict resolution unit 113, and inputs the control values obtained from the respective neural networks (5, 6) to the input layer 71 of the third neural network 7. Then, the control unit 11 determines the firing of each neuron included in each layer 71 to 73 in the forward propagation direction, so that the control has been modified so as to eliminate the conflict from the output layer 73 of the third neural network 7. Get the value. Thereby, in the present embodiment, when a conflict occurs in the control of the air conditioner 4 using each neural network (5, 6), the control of the air conditioner 4 can be modified so as to eliminate the conflict. ..

Here, in this step S103, each control value is input to the third neural network 7 without distinguishing whether or not a conflict occurs in the control of the air conditioner 4 depending on each control value acquired in step S102. ing. That is, even when there is no conflict in the control of the air conditioner 4 due to the control values acquired in step S102, the control values acquired in step S102 are input to the third neural network 7.

At this time, the third neural network 7 may be learned to output each control value as it is when the control values of the air conditioner 4 do not conflict with each control value acquired in step S102, or the conflict may occur. May be trained to output the corrected control value as in the case of occurrence. In the following, the control value output as it is from the third neural network 7 without modifying the input control value is also referred to as "corrected control value (corrected control value)".

(Step S104)
In the next step S104, the control unit 11 controls the operation of the air conditioner 4 based on the corrected control value acquired from the third neural network 7 in the step S103. The control value indicates, for example, the desired room temperature to be achieved by operating the air conditioner 4. The control unit 11 compares the desired room temperature indicated by the control value with the current room temperature, and controls the operation of cooling and heating of the air conditioner 4 so as to reach the desired room temperature.

As described above, the control unit 11 ends the process related to this operation example. The control unit 11 may repeatedly execute the above steps S101 to S104 periodically or irregularly. As a result, the control device 1 can continuously control the operation of the air conditioner 4 according to the preference of each user (A, B).

[Control device for data collection]
Next, an operation example of the data collection control device 2 will be described with reference to FIG. FIG. 9 is a flowchart illustrating an example of the processing procedure of the data collection control device 2. The processing procedure described below corresponds to the "learning data creation method" of the present invention. However, the processing procedure described below is only an example, and each processing may be changed as much as possible. Further, with respect to the processing procedure described below, steps can be omitted, replaced, and added as appropriate according to the embodiment.

(Start-up)
In the same manner as described above, the system including the data collection control device 2 and the air conditioner 4 is appropriately started. When the system is started, the data collection control device 2 reads the data collection control program 221 and executes the initial setting process. That is, the control unit 21 sets the structure of each neural network (5, 6), the weight of the connection between each neuron, and the threshold value of each neuron with reference to each learning result data (122, 123). Then, the control unit 21 creates learning data 223 for constructing the third neural network 7 according to the following processing procedure.

(Steps S201 and S202)
In step S201, the control unit 21 acquires various information to be input to each neural network (5, 6) from each user (A, B) in the same manner as in step S101.

In the next step S202, the control unit 21 functions as the first control processing unit 211 as in the above step S102, and inputs various information acquired from the user A to the input layer 51 of the first neural network 5. do. Then, the control unit 21 determines the firing of each neuron included in each layer 51 to 53 in the forward propagation direction, and according to the preference of the user A output from the output layer 53 of the first neural network 5. Acquire the control value for the air conditioner 4.

Further, the control unit 21 functions as a second control processing unit 212, and inputs various information acquired from the user B to the input layer 61 of the second neural network 6. Then, the control unit 21 determines the firing of each neuron included in each layer 61 to 63 in the forward propagation direction, and according to the preference of the user B output from the output layer 63 of the second neural network 6. Acquire the control value for the air conditioner 4.

(Steps S203 and S204)
In the next step S203, the control unit 21 controls the operation of the air conditioner 4 which is the control target device based on each control value acquired from each neural network (5, 6) in the step S202. Then, in step S204, the control unit 21 determines whether or not a conflict occurs in the control of the air conditioner 4.

At this time, the control unit 21 may actually operate the air conditioner 4 in step S203 to determine whether or not each control value acquired from each neural network (5, 6) causes a conflict. Further, the control unit 21 does not actually operate the air conditioner 4 in step S203, but simulates the operation of the air conditioner 4 based on each control value acquired from each neural network (5, 6). , It may be determined whether or not a conflict occurs.

The method for determining whether or not a conflict occurs may be appropriately set according to the embodiment. For example, if the control of the air conditioner 4 based on the control values acquired from each neural network (5, 6) cannot be executed at the same time, the control unit 21 may determine that a conflict occurs in the control of the air conditioner 4. good. When it is determined that a conflict occurs in the control of the air conditioner 4, the control unit 21 proceeds to the next step S205. On the other hand, when it is determined that there is no conflict in the control of the air conditioner 4, the control unit 21 ends the process according to this operation example.

(Step S205)
In the next step S205, the control unit 21 functions as a correction value determination unit 213, and of each control value acquired from each neural network (5, 6) so as to eliminate the conflict caused in the control of the air conditioner 4. Determine the correction value. As a result, the control unit 21 acquires the corrected control value so as to eliminate the conflict.

The method for modifying the control value may be appropriately selected according to the embodiment. For example, the control unit 21 may determine the correction value according to a predetermined rule. In this case, when it is stipulated as a predetermined rule to give priority to either one of the users A and B, the control unit 21 receives one of the control values acquired from each neural network (5, 6). Give priority to. That is, the control unit 21 treats the priority control value as the corrected control value. Further, when it is stipulated as a predetermined rule that both users A and B should be treated equally, the control unit 21 corrects by averaging the control values acquired from each neural network (5, 6). Get the completed control value.

It should be noted that the predetermined rule does not have to be limited to such an example. For example, when the priority is set for each of the users A and B as a predetermined rule, the control unit 21 has corrected the weighted average of the control values acquired from each neural network (5, 6). It may be acquired as a control value.

Further, for example, the control unit 21 may accept the input of the correction value from the operator. That is, the control unit 21 may determine the corrected control value based on the input from the operator. In this case, the data collection control device 2 may be connected to an input device such as a keyboard or a microphone via the external interface 23. As a result, the operator can input the corrected control value by keyboard input, voice input, or the like.

(Step S206)
In the next step S206, the control unit 21 functions as a learning data creation unit 214, and obtains the control value acquired from each neural network (5, 6) in step S202 and the corrected control value determined in step S205. Make a pair. As a result, the control unit 21 creates learning data 223 in which each control value before modification is input data and the modified control value is teacher data. Then, the control unit 21 stores the created learning data 223 in the storage unit 22.

As described above, the control unit 21 ends the process related to this operation example. The control unit 21 can collect a plurality of learning data 223s by repeatedly executing the series of processes of steps S201 to S206.

When constructing a third neural network 7 that outputs each control value as it is when there is no conflict in the control of the air conditioner 4, the control unit 21 determines in step S204 that no conflict occurs. Learning data 223 may be created by using each control value as input data and teacher data.

Further, when constructing the third neural network 7 that modifies the control values obtained from the respective neural networks (5, 6) even when there is no conflict in the control of the air conditioner 4, the control unit 21 performs the control unit 21 in step S204. Even when it is determined that no conflict occurs, the processes of steps S205 and S206 may be executed.

In this case, the control unit 21 may accept the input from the operator and determine the modified value so that the modified control value suitable for the preferences of both users A and B can be obtained. As a result, when the control values obtained from each neural network (5, 6) are input, it is used for the construction of the third neural network 7 that outputs the corrected control values suitable for the preferences of both users A and B. Possible learning data 223 can be created.

[Learning device]
Next, an operation example of the learning device 3 will be described with reference to FIG. FIG. 10 is a flowchart illustrating an example of the processing procedure of the learning device 3. The processing procedure described below is only an example, and each processing may be changed as much as possible. Further, with respect to the processing procedure described below, steps can be omitted, replaced, and added as appropriate according to the embodiment.

(Step S301)
In step S301, the control unit 31 functions as the learning data acquisition unit 311 and acquires the learning data 223 created by the data collection control device 2.

The method of transferring the learning data 223 created by the data collection control device 2 to the learning device 3 may be appropriately selected depending on the embodiment. For example, when the learning device 3 and the data collection control device 2 are connected via a network, the control unit 31 accesses the data collection control device 2 via the network to obtain the learning data 223. Can be obtained. Further, for example, the learning data 223 created by the data collection control device 2 may be stored in another information processing device (storage device) such as NAS (Network Attached Storage). In this case, the control unit 31 can acquire the learning data 223 by accessing the other information processing device. Further, for example, the learning data 223 created by the data collection control device 2 may be stored in the storage medium 91. In this case, the control unit 31 can acquire the learning data 223 from the storage medium 91 via the drive 36. The number of learning data 223 acquired in this step S301 may be appropriately determined according to the embodiment so that the learning neural network 8 can be learned.

(Step S302)
In the next step S302, the control unit 31 functions as a learning processing unit 312, and when the control values obtained from each neural network (5, 6) are input using the learning data 223 acquired in step S301, a conflict occurs. The training neural network 8 is trained so as to output the corrected control value so as to be eliminated.

Specifically, first, the control unit 31 prepares a learning neural network 8 to be subjected to learning processing. The configuration of the neural network 8 to be prepared, the initial value of the weight of the connection between each neuron, and the initial value of the threshold value of each neuron may be given by the template or by the input of the operator. Further, when performing re-learning, the control unit 31 may prepare a neural network 8 for learning based on the conflict resolution learning result data 124 to be re-learned.

Next, the control unit 31 uses the control values obtained from the neural networks (5, 6) included in the learning data 223 acquired in step S301 as input data, and the corrected control values as teacher data, as the neural network. 8 learning is performed. A gradient descent method, a stochastic gradient descent method, or the like may be used for learning the neural network 8.

For example, the control unit 31 inputs the control values obtained from each neural network (5, 6) included in the learning data 223 into the input layer 81, and performs arithmetic processing in the forward propagation direction of the training neural network 8. conduct. As a result, the control unit 31 obtains an output value from the output layer 83 of the learning neural network 8. Next, the control unit 31 calculates an error between the output value output from the output layer 83 and the corrected control value included in the learning data 223. Subsequently, the control unit 31 calculates the error of the connection weight between each neuron and the error of each threshold value of each neuron by using the error of the output value calculated by the error back propagation method. Then, the control unit 31 updates the weight of the connection between each neuron and the value of each threshold value of each neuron based on each calculated error.

The control unit 31 learns the neural network 8 by repeating this series of processes for each learning data 223 until the output value output from the output layer 83 matches the corresponding corrected control value. .. As a result, when the control values obtained from each neural network (5, 6) are input, the neural network 8 that outputs the corrected control value so as to eliminate the conflict can be constructed.

(Step S303)
In the next step S303, the control unit 31 functions as a learning processing unit 312 to provide information indicating the configuration of the constructed neural network 8, the weight of the connection between each neuron, and the threshold value of each neuron as conflict resolution learning result data. It is stored in the storage unit 32 as 124. As a result, the control unit 31 ends the learning process related to this operation example.

When the learning device 3 and the control device 1 can be connected via a network, the control unit 31 appropriately transfers the created conflict resolution learning result data 124 to the control device 1 after the process of step S303 is completed. You may. Further, the control unit 31 may update the conflict resolution learning result data 124 periodically or irregularly by executing the learning processes of steps S301 to S303 periodically or irregularly. Then, the control unit 31 transfers the created conflict resolution learning result data 124 to the control device 1 each time the learning process is executed, so that the conflict resolution learning result data 124 held by the control device 1 is periodically or not. It may be updated regularly.

[Action / Effect]
As described above, in the present embodiment, the operation of the air conditioner 4 can be controlled by each neural network (5, 6) so as to suit the preference of each user (A, B). However, if conflicting instructions are given by both parties, a conflict may occur in the control of the air conditioner 4. On the other hand, in the present embodiment, the control values obtained from the respective neural networks (5, 6) are modified so as to eliminate the conflict by the third neural network 7 used in the process of step S103. can do. Therefore, according to the present embodiment, it is possible to prevent the air conditioner 4 from becoming inoperable even if a conflict occurs between the controls by each user (A, B).

§4 Modifications Although the embodiments of the present invention have been described in detail above, the above description is merely an example of the present invention in all respects. Needless to say, various improvements and modifications can be made without departing from the scope of the present invention. For example, the following changes can be made. In the following, the same reference numerals will be used for the same components as those in the above embodiment, and the same points as in the above embodiment will be omitted as appropriate. The following modifications can be combined as appropriate.

<4.1>
In the above embodiment, an air conditioner is exemplified as a controlled object device controlled by the control device 1. However, the type of the device to be controlled does not have to be limited to the air conditioner, and may be appropriately selected according to the embodiment. The control target device may be, for example, a robot device or the like.

Further, in the above embodiment, the first neural network 5 and the second neural network 6 control the same controlled object device (air conditioner 4). However, the controlled object device controlled by the first neural network 5 and the controlled object device controlled by the second neural network 6 may be different from each other.

For example, in the first neural network 5, the first robot device may be the control target as the first control target device. Then, the second neural network 6 may use a second robot device different from the first robot device as a control target as the second control target device. In this case, for example, when a control command for moving to the same position at the same timing is issued to the first robot device and the second robot device, a conflict may occur between the controls of both robot devices.

Further, in the above embodiment, two learning devices (first neural network 5 and second neural network 6) are used as learning devices that issue control values for controlling the operation of the controlled device. However, the number of learners that issue control values that control the operation of the controlled device is not limited to two, and may be three or more.

<4.2>
Regarding the specific hardware configurations of the control device 1, the data collection control device 2, and the learning device 3, components can be omitted, replaced, or added as appropriate according to the embodiment. For example, the control unit 11 may include a plurality of processors. The control device 1 and the data collection control device 2 may be configured to include a communication interface and enable data exchange with other information processing devices via a network. The control device 1, the data collection control device 2, and the learning device 3 may each be composed of a plurality of computers.

Further, each of the control device 1 and the data collection control device 2 depends on an information processing device such as an ECU (Electronic Control Unit) designed exclusively for the provided service, and a control target device to be controlled. A general-purpose desktop PC, tablet PC, mobile phone, or the like may be used as appropriate. Further, as the learning device 3, a general-purpose server device, a desktop PC, or the like may be used in addition to the information processing device designed exclusively for the provided service.

<4.3>
Further, in the above embodiment, as shown in FIGS. 5 to 7, a general feedforward neural network having a multi-layer structure is used as each neural network 5 to 8. However, the types of each neural network 5 to 8 do not have to be limited to such an example, and may be appropriately selected depending on the embodiment. For example, when an image is used as input data, a convolutional neural network including a convolutional layer and a pooling layer may be used for each neural network 5 to 8. Further, for example, when time-series data is used as input data, a recursive neural network having a recursive connection from the output side to the input side, such as an intermediate layer to an input layer, is used for each neural network 5 to 8. You may. The number of layers of each neural network 5 to 8, the number of neurons in each layer, the connection relationship between neurons, and the transfer function of each neuron may be appropriately determined according to the embodiment.

<4.4>
Further, in step S103, the control values acquired from each neural network (5, 6) in step S102 are transferred to the third neural network 7 without distinguishing whether or not a conflict occurs in the control of the air conditioner 4. I'm typing. However, the processing procedure of the control device 1 does not have to be limited to such an example. Only when the control value acquired from each neural network (5, 6) in step S102 causes a conflict in the control of the air conditioner 4, the control unit 11 uses the control value acquired in step S102 as the third neural network. You may enter in 7. In this case, when the control values acquired from the neural networks (5, 6) in step S102 do not cause a conflict in the control of the air conditioner 4, the control unit 11 omits the step S103 and skips the step S103 to the next step S104. By executing the process, the control values acquired from each neural network (5, 6) may be used as they are for the control of the air conditioner 4.

Further, in step S103, the third neural network 7 is used to eliminate the conflict that occurs in the control of the air conditioner 4. However, the method for resolving the conflict that occurs in the control of the air conditioner 4 does not have to be limited to such an example. Conflicts that occur in the control of the air conditioner 4 may be resolved without using a neural network.

FIG. 11 schematically illustrates the control device 1A according to this modification. The control device 1A is configured in the same manner as the control device 1 except that it does not hold the conflict resolution learning result data 124 and includes a conflict resolution unit 113A that does not use a neural network. In this case, the control unit 11 functions as the conflict resolution unit 113A in step S103, and determines whether or not a conflict occurs in the control of the air conditioner 4 based on the control values acquired from each neural network (5, 6). do. For example, the control unit 11 determines whether or not a conflict occurs in the control of the air conditioner 4 by simulating the operation of the air conditioner 4.

Then, when it is determined that there is no conflict in the control of the air conditioner 4, the control unit 11 controls the operation of the air conditioner 4 based on the control values acquired from each neural network (5, 6). On the other hand, when it is determined that a conflict occurs in the control of the air conditioner 4, the control unit 11 corrects the control value acquired from each neural network (5, 6) so as to eliminate the conflict, and the corrected control. The operation of the air conditioner 4 is controlled based on the value.

The method for modifying the control value may be appropriately selected according to the embodiment, as in step S205. For example, the control unit 11 may determine the corrected correction value by giving priority to any one of the control values acquired from each neural network (5, 6). Further, for example, the control unit 11 may determine the corrected control value by averaging the control values acquired from each neural network (5, 6).

<4.5>
Further, in the above embodiment, the control unit 11 uses the control value acquired from each neural network (5, 6) as the third neural network without specifying what kind of competition is generated in the control of the air conditioner 4. Input to network 7. However, the processing procedure of the control device 1 does not have to be limited to such an example, and the control unit 11 determines what kind of control value the control value acquired from each neural network (5, 6) is for controlling the air conditioner 4. You may specify whether to cause a conflict.

FIG. 12 schematically illustrates the control device 1B according to this modification. The control device 1B includes a competition type identification unit 114 that specifies competition type information 125 indicating how the control of the air conditioner 4 competes based on the control values acquired from each neural network (5, 6). It is configured in the same manner as the control device 1 except that the specified competition type information 125 is used for the input of the third neural network 7.

In this case, the control unit 11 shows how the controls of the air conditioner 4 compete with each other based on the control values acquired from each neural network (5, 6) before executing the step S103. Identify information 125. The type (method) of competition may be appropriately set according to the embodiment. Then, in step S103, the control unit 11 inputs the control value and the competition type information 125 acquired from each neural network (5, 6) to the input layer 71 of the third neural network 7.

As a result, the control device 1B can surely change the method of correcting the control value according to the type of competition. For example, when the user A desires a room temperature higher than that of the user B, the third neural network 7 sets the average value of the control values acquired from each neural network (5, 6) as the corrected control value. It may be output as. Then, when the user B desires a room temperature higher than that of the user A, the third neural network 7 preferentially outputs the control value acquired from the neural network 5 as the corrected control value. You may.

As illustrated in FIG. 13, a learning device such as a neural network may be used to identify how the controls of the air conditioner 4 compete with each other. FIG. 13 schematically illustrates the control device 1C according to this modification. The control device 1C is configured in the same manner as the control device 1B, except that the control device 1C includes a competition type identification unit 114C that specifies the competition type information 125 by using the fourth neural network 115. The fourth neural network 115 has been learned to output the output value corresponding to the conflict type information 125 when the control value output from each neural network (5, 6) is input. The fourth neural network 115 may be configured in the same manner as each neural network 5 to 7, for example. In the above processing procedure, when the system is started, the control unit 11 refers to the learning result data 126 to set the structure of the fourth neural network 115, the weight of the connection between each neuron, and the threshold value of each neuron. conduct. As a result, the conflicts that occur in the control of the air conditioner 4 can be classified in a complicated manner, and an appropriate resolving method can be adopted for each classification.

<4.6>
Further, in the above embodiment (and modification), each learner is configured by a neural network. However, the type of each learner does not have to be limited to the neural network, and may be appropriately selected according to the embodiment. For each learning device, for example, a support vector machine, a self-organizing map, a learning device that learns by reinforcement learning, or the like may be used.

<4.7>
Further, a learning device for creating each of the above neural networks (5, 6) may be prepared. For example, by changing the learning data used for machine learning from the learning data 223 to learning data for learning the control suitable for each user (A, B), each of the learned data by the learning device 3 A neural network (5, 6) can be created. The learning data for learning the control suitable for each user (A, B) includes various information that is input data acquired from each user (A, B) and the preference of each user (A, B). It can be created by combining with the original control value which is the teacher data suitable for. The learning device uses such learning data to execute the processes of steps S301 to 303 to construct each trained neural network (5, 6), and each motion control learning result data (122). , 123) can be created.

Similarly, a learning device for creating the fourth neural network 115 may be prepared. For example, by changing the learning data used for machine learning from the learning data 223 to learning data for learning the identification of the type of competition, the learning device 3 creates a learned fourth neural network 115. can do. The learning data for learning the identification of the type of conflict is a combination of a control value as input data acquired from each neural network (5, 6) and an output value corresponding to the contention type information 125 as teacher data. It can be created by. The learning device can construct the trained fourth neural network 115 and create the learning result data 126 by executing the processes of steps S301 to 303 using such learning data. ..

1.1A, 1B, 1C ... Control device,
11 ... Control unit, 12 ... Storage unit, 13 ... External interface,
111 ... 1st control processing unit, 112 ... 2nd control processing unit,
113 / 113A ... Conflict Resolution Department,
114 / 114C ... Competitive type identification part,
115 ... Fourth neural network,
121 ... control program, 122 ... first motion control learning result data,
123 ... Second motion control learning result data,
124 ... Conflict resolution learning result data,
125 ... Competitive type information, 126 ... Learning result data,
2 ... Data collection control device,
21 ... Control unit, 22 ... Storage unit, 23 ... External interface,
211 ... 1st control processing unit, 212 ... 2nd control processing unit,
213 ... Correction value determination unit, 214 ... Learning data creation unit,
221 ... Data collection control program, 223 ... Learning data,
3 ... Learning device,
31 ... control device, 32 ... storage unit, 33 ... communication interface,
34 ... Input device, 35 ... Output device, 36 ... Drive,
311 ... Learning data acquisition unit 312 ... Learning processing unit,
321 ... Learning program,
5 ... First neural network,
51 ... Input layer, 52 ... Intermediate layer (hidden layer), 53 ... Output layer,
6 ... Second neural network,
61 ... Input layer, 62 ... Intermediate layer (hidden layer), 63 ... Output layer,
7 ... Third neural network,
71 ... Input layer, 72 ... Intermediate layer (hidden layer), 73 ... Output layer,
8 ... Neural network for learning,
81 ... Input layer, 82 ... Intermediate layer (hidden layer), 83 ... Output layer,
91 ... Storage medium

Claims (7)

A first control target device that controls the operation of the first control target device based on a control value output from a learned first learner that has been trained to control the operation of the first control target device. Control processing unit and
A second control target device that controls the operation of the second control target device based on a control value output from the learned second learner that has been trained to control the operation of the second control target device. Control processing unit and
The control of the first control target device based on the control value output from the first learner and the control of the second control target device based on the control value output from the second learner conflict with each other. In this case, when the control value of the first control target device output from the first learner and the control value of the second control target device output from the second learner are input, the control value of the second control target device is input. Using the trained third learner that has been trained to output the control value of the first control target device and the control value of the second control target device modified to eliminate the conflict. And the conflict resolution department that resolves the conflict,
To prepare
Control device. The first control is based on the control value of the first control target device output from the first learner and the control value of the second control target device output from the second learner. Further provided with a conflict type identification unit that specifies conflict type information indicating how the control of the target device and the second control target device conflicts with each other.
The conflict resolving unit further inputs the specified conflict type information to the third learner.
The control device according to claim 1. The competition type identification unit inputs the control value of the first control target device output from the first learner and the control value of the second control target device output from the second learner. Then, the conflict type information is specified by using the learned fourth learner that has been trained to output the output value corresponding to the conflict type information.
The control device according to claim 2. The first, second, third and fourth learners are each composed of a neural network.
The control device according to claim 3. The conflict resolving unit controls the first controlled device based on the control value output from the first learner, and the second control based on the control value output from the second learner. By giving priority to one of the controls of the target device, the conflict is resolved.
The control device according to any one of claims 1 to 4. The first control target device and the second control target device are the same control target device, and the first control target device and the second control target device are the same control target device.
The conflict resolving unit resolves the conflict by averaging the control value output from the first learner and the control value output from the second learner.
The control device according to any one of claims 1 to 5. To the computer that controls the operation of the first controlled device and the second controlled device,
A step of acquiring a control value for controlling the first control target device output from the learned first learner that has learned to control the operation of the first control target device, and a step of acquiring the control value for controlling the first control target device.
A step of acquiring a control value for controlling the second control target device output from the learned second learner that has learned to control the operation of the second control target device, and a step of acquiring the control value for controlling the second control target device.
The control of the first control target device based on the control value output from the first learner and the control of the second control target device based on the control value output from the second learner conflict with each other. In this case, when the control value of the first control target device output from the first learner and the control value of the second control target device output from the second learner are input, the control value of the second control target device is input. Modified from the trained third learner that has been trained to output the control value of the first control target device and the control value of the second control target device modified to eliminate the conflict. The step of acquiring the control value of the first control target device and the control value of the second control target device, and
A step of controlling the first control target device and the second control target device based on the acquired control value, and
A control program for executing.

Images