+

WO2018211927A1 - Appareil de commande, programme de commande, procédé de création de données d'apprentissage et procédé d'apprentissage - Google Patents

Appareil de commande, programme de commande, procédé de création de données d'apprentissage et procédé d'apprentissage Download PDF

Info

Publication number
WO2018211927A1
WO2018211927A1 PCT/JP2018/016703 JP2018016703W WO2018211927A1 WO 2018211927 A1 WO2018211927 A1 WO 2018211927A1 JP 2018016703 W JP2018016703 W JP 2018016703W WO 2018211927 A1 WO2018211927 A1 WO 2018211927A1
Authority
WO
WIPO (PCT)
Prior art keywords
control
learning
target apparatus
control target
conflict
Prior art date
Application number
PCT/JP2018/016703
Other languages
English (en)
Inventor
Tanichi Ando
Original Assignee
Omron Corporation
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Omron Corporation filed Critical Omron Corporation
Publication of WO2018211927A1 publication Critical patent/WO2018211927A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks

Definitions

  • the present invention relates to a control apparatus, a control program, a learning data creation method and a learning method.
  • JP 2016-532355A proposes an intelligent housing system that utilizes a neural apparatus. Specifically, JP 2016-532355A proposes a system that performs machine learning for learning a function of performing adjustment control of an electronic apparatus in an operating state commonly used by a user, using identity information, coordinate information and control information from when the user adjusted the electronic apparatus in the past.
  • conflict could possibly arise between a plurality of controls directed to one or a plurality of control target apparatuses.
  • conflict will arise in control of the control target apparatus when control commands for executing different operations are issued from the learning machines of different users.
  • conflict will arise in control of the plurality of control target apparatuses when control commands for executing operations that cannot be executed at the same time such as moving to the same location are issued.
  • the inventor found that, with conventional systems, there was a problem in that the control target apparatuses could possibly become inoperative.
  • the present invention was made in view of such situations, and an object thereof is to provide a technology for ensuring that, even when conflict arises between a plurality of controls, control target apparatuses do not become inoperative.
  • the present invention employs the following configurations, in order to solve the abovementioned problems.
  • a control apparatus configured with a first control processing unit configured to control operation of a first control target apparatus, based on a control value output from a trained first learning machine that has performed learning for controlling the operation of the first control target apparatus, a second control processing unit configured to control operation of a second control target apparatus, based on a control value output from a trained second learning machine that has performed learning for controlling the operation of the second control target apparatus, and a conflict resolution unit configured to, in a case where control of the first control target apparatus that is based on the control value output from the first learning machine conflicts with control of the second control target apparatus that is based on the control value output from the second learning machine, resolve the conflict by correcting control of the first control target apparatus and the second control target apparatus.
  • the first control processing unit controls the operation of the first control target apparatus, utilizing the first learning machine.
  • the second control processing unit controls the operation of the second control target apparatus, utilizing the second learning machine.
  • the conflict resolution unit resolves the conflict, by correcting control of the first control target apparatus and the second control target apparatus. Accordingly, with this configuration, it is possible to ensure that, even when conflict arises between a plurality of controls, control target apparatuses do not become inoperative.
  • the first control target apparatus and the second control target apparatus may be the same control target apparatus, or may be mutually different control target apparatuses.
  • the control target apparatus may include all types of apparatuses that can be targeted for control, and may, for example, be an air conditioning apparatus, a robotic apparatus or the like.
  • the conflict resolution unit may resolve the conflict, utilizing a trained third learning machine that has learned to output a control value of the first control target apparatus and a control value of the second control target apparatus corrected so as to resolve the conflict, when the control value of the first control target apparatus output from the first learning machine and the control value of the second control target apparatus output from the second learning machine are input.
  • the control apparatus may be further provided with a conflict type specification unit configured to specify conflict type information indicating how control of the first control target apparatus and control of the second control target apparatus conflict, based on the control value of the first control target apparatus output from the first learning machine and the control value of the second control target apparatus output from the second learning machine, and the conflict resolution unit may further input the specified conflict type information to the third learning machine.
  • a suitable conflict resolution method can be employed according to the manner (type) of conflict.
  • the conflict type specification unit may specify the conflict type information, utilizing a trained fourth learning machine that has learned to output an output value corresponding to the conflict type information, when the control value of the first control target apparatus output from the first learning machine and the control value of the second control target apparatus output from the second learning machine are input.
  • the first, second, third and fourth learning machines may be respectively constituted by a neural network.
  • the conflict resolution unit may resolve the conflict, by prioritizing one of control of the first control target apparatus that is based on the control value output from the first learning machine and control of the second control target apparatus that is based on the control value output from the second learning machine.
  • the first control target apparatus and the second control target apparatus may be the same control target apparatus, and the conflict resolution unit may resolve the conflict, by averaging the control value output from the first learning machine and the control value output from the second learning machine.
  • control apparatus includes an information processing method that realizes the above configurations, a computer program, or a storage medium, readable by a computer or other apparatus, machine or the like, on which such a computer program is recorded.
  • the storage medium that is readable by a computer or the like is a medium that stores information such as programs and the like through an electrical, magnetic, optical, mechanical or chemical operation.
  • a control program is for causing a computer that controls operation of a first control target apparatus and a second control target apparatus to execute a step of acquiring a control value for controlling the first control target apparatus output from a trained first learning machine that has performed learning for controlling the operation of the first control target apparatus, a step of acquiring a control value for controlling the second control target apparatus output from a trained second learning machine that has performed learning for controlling the operation of the second control target apparatus, a step of acquiring, in a case where control of the first control target apparatus that is based on the control value output from the first learning machine conflicts with control of the second control target apparatus that is based on the control value output from the second learning machine, a control value of the first control target apparatus and a control value of the second control target apparatus corrected so as to resolve the conflict, and a step of controlling the first control target apparatus and the second control target apparatus based on the acquired control values.
  • a learning data creation method is provided with a step of acquiring a control value for controlling a first control target apparatus output from a trained first learning machine that has performed learning for controlling operation of the first control target apparatus, a step of acquiring a control value for controlling a second control target apparatus output from a trained second learning machine that has performed learning for controlling operation of the second control target apparatus, a step of determining whether control of the first control target apparatus that is based on the control value output from the first learning machine conflicts with control of the second control target apparatus that is based on the control value output from the second learning machine, a step of determining, in a case where control of the first control target apparatus that is based on the control value output from the first learning machine conflicts with control of the second control target apparatus that is based on the control value output from the second learning machine, a correction value of the control values of the first control target apparatus and the second control target apparatus so as to resolve the conflict, and a step of creating learning data for performing learning of a learning machine, with the control
  • the correction value may be determined by input from an operator.
  • the control target apparatus is an apparatus that is utilized by a person
  • optimal learning data for building the third learning machine that is utilized in resolving conflict that can occur in control of control target apparatuses can be created.
  • the correction value may determined in accordance with a predetermined rule.
  • a learning method is provided with a step of acquiring the learning data created by the learning data creation method according to any of the above forms, and a step of performing learning of a learning machine with the acquired learning data.
  • the third learning machine that is utilized in resolving conflict that can occur in control of control target apparatuses can be built.
  • a technology can be provided for ensuring that, even when conflict arises between a plurality of controls, control target apparatuses do not become inoperative.
  • Fig. 1 schematically illustrates an example of a scenario in which a control apparatus and a learning apparatus according to an embodiment are applied.
  • Fig. 2 schematically illustrates an example of a hardware configuration of the control apparatus according to the embodiment.
  • Fig. 3 schematically illustrates an example of a hardware configuration of a data collection control apparatus according to the embodiment.
  • Fig. 4 schematically illustrates an example of a hardware configuration of the learning apparatus according to the embodiment.
  • Fig. 5 schematically illustrates an example of a functional configuration of the control apparatus according to the embodiment.
  • Fig. 6 schematically illustrates an example of a functional configuration of the data collection control apparatus according to the embodiment.
  • Fig. 7 schematically illustrates an example of a functional configuration of the learning apparatus according to the embodiment.
  • Fig. 1 schematically illustrates an example of a scenario in which a control apparatus and a learning apparatus according to an embodiment are applied.
  • Fig. 2 schematically illustrates an example of a hardware configuration of the control apparatus according to the embodiment.
  • FIG. 8 illustrates an example of a processing procedure of the control apparatus according to the embodiment.
  • Fig. 9 illustrates an example of a processing procedure of the data collection control apparatus according to the embodiment.
  • Fig. 10 illustrates an example of a processing procedure of the learning apparatus according to the embodiment.
  • Fig. 11 illustrates an example of the configuration of a control apparatus according to a variation.
  • Fig. 12 illustrates an example of the configuration of a control apparatus according to a variation.
  • Fig. 13 illustrates an example of the configuration of a control apparatus according to a variation.
  • FIG. 1 schematically illustrates an example of a scenario in which a control apparatus 1 and a learning apparatus 3 according to the present embodiment are applied.
  • the control apparatus 1 is an information processing apparatus that controls the operation of an air conditioning apparatus 4 which is a control target apparatus, in accordance with instructions from a plurality of users (users A and B in Fig. 1).
  • the air conditioning apparatus 4 is, for example, a well-known air conditioner that adjusts the indoor temperature, and corresponds to "a first control target apparatus" and "a second control target apparatus" of the present invention. That is, in the present embodiment, the first control target apparatus and the second control target apparatus are the same.
  • the first control target apparatus and the second control target apparatus need not, however, be limited to such an example, and may be different apparatuses.
  • the control apparatus 1 is provided with two learning machines for controlling the operation of the air conditioning apparatus 4.
  • the first learning machine first neural network 5 discussed later
  • the second learning machine second neural network 6 discussed later
  • the control apparatus 1 controls the operation of the air conditioning apparatus 4, based on control values that are respectively output from the first learning machine and the second learning machine.
  • conflict could possibly occur in control of the operation of the air conditioning apparatus 4.
  • the control value that is output from the first learning machine constitutes a command for setting the room temperature to 26 degrees
  • the control value that is output from the second learning machine constitutes a command for setting the room temperature to 22 degrees in a situation where the room temperature is 24 degrees
  • conflict occurs in control of the operation of the air conditioning apparatus 4.
  • control apparatus 1 in the case where control of the air conditioning apparatus 4 that is based on the control value output from the first learning machine conflicts with control of the air conditioning apparatus 4 that is based on the control value output from the second learning machine, resolves the conflict by correcting control of the air conditioning apparatus 4. Specifically, the control apparatus 1 according to the present embodiment resolves conflict of controls of the air conditioning apparatus 4, utilizing a third learning machine (third neural network 7 discussed later).
  • the third learning machine has learned to output a control value corrected so as to resolve conflict (hereinafter, also referred to as “the corrected control value”), when control values that are respectively output from the first learning machine and the second learning machine are input.
  • the control apparatus 1 is able to obtain a control value corrected such that conflict does not occur, by inputting the control values that are respectively obtained from the first learning machine and the second learning machine to the third learning machine.
  • the control apparatus 1 controls the operation of the air conditioning apparatus 4, based on the corrected control value obtained in this way.
  • the learning apparatus 3 is an information processing apparatus that performs machine learning of the third learning machine.
  • the learning apparatus 3 according to the present embodiment collects learning data to be utilized in machine learning of the third learning machine, using a data collection control apparatus 2.
  • the data collection control apparatus 2 similarly to the control apparatus 1, controls the operation of the air conditioning apparatus 4 to suit the preferences of the users (A, B), utilizing the first learning machine and the second learning machine.
  • the data collection control apparatus 2 is, however, dissimilar to the control apparatus 1 in that conflict of controls of the air conditioning apparatus 4 is not resolved (third learning machine is not utilized).
  • the data collection control apparatus 2 determines whether control of the air conditioning apparatus 4 that is based on the control value obtained from the first learning machine conflicts with control of the air conditioning apparatus 4 that is based on the control value obtained from the second learning machine. In the case where it is determined that control of the air conditioning apparatus 4 that is based on the control value obtained from the first learning machine conflicts with control of the air conditioning apparatus 4 that is based on the control value obtained from the second learning machine, the data collection control apparatus 2 determines a correction value of the control values so as to resolve the conflict.
  • the data collection control apparatus 2 prioritizes one of the control values that are respectively obtained from the first learning machine and the second learning machine. That is, the data collection control apparatus 2 takes the prioritized control value as the corrected control value. Also, for example, the data collection control apparatus 2 calculates an average value of the control values that are respectively obtained from the first learning machine and the second learning machine as the corrected control value. The data collection control apparatus 2 is thereby able to acquire a corrected control value determined so as to resolve the above conflict.
  • the data collection control apparatus 2 then creates learning data to be utilized in machine learning of the third learning machine, with the control values respectively obtained from the first learning machine and the second learning machine as input data, and with the corrected control value that is obtained as described above as training data.
  • the data collection control apparatus 2 creates learning data, by pairing the control values before correction with the corrected control value.
  • the learning apparatus 3 builds a trained third learning machine that is utilizable in the above control apparatus 1, by acquiring the learning data created in this way, and performing machine learning of the third learning machine using the acquired learning data.
  • the control apparatus 1 may, for example, acquire the trained third learning machine from the learning apparatus 3, via a network.
  • the trained third learning machine may be embedded in the control apparatus 1 as embedded data.
  • the control apparatus 1 is able to control the operation of the air conditioning apparatus 4 to suit the preferences of the users (A, B), by using the trained first learning machine and second learning machine.
  • the conflict can be resolved utilizing the third learning machine. Accordingly, with the present embodiment, it is possible to ensure that, even when conflict arises between controls by the respective users (A, B), the air conditioning apparatus 4 does not become inoperative.
  • FIG. 2 schematically illustrates an example of the hardware configuration of the control apparatus 1 according to the present embodiment.
  • the control apparatus 1 is a computer to which a control unit 11, a storage unit 12 and an external interface 13 are electrically connected.
  • 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 stores a control program 121 that is executed with the control unit 11, first operation control learning result data 122 that indicates information relating to the trained first learning machine, second operation control learning result data 123 that indicates information relating to the trained second learning machine, conflict resolution learning result data 124 that indicates information relating to the trained third learning machine, and the like.
  • the control program 121 is a program for causing the control unit 11 to execute processing (Fig. 8) for controlling the operation of the air conditioning apparatus 4 which will be discussed later.
  • the first operation control learning result data 122 is data that is utilized in setting the trained first learning machine.
  • the second operation control learning result data 123 is data that is utilized in setting the trained second learning machine.
  • the conflict resolution learning result data 124 is data that is utilized in setting the trained third learning machine. A detailed description thereof will be given later.
  • the external interface 13 is an interface for connecting to an external apparatus, and is configured as appropriate according to the external apparatus to be connected.
  • the control apparatus 1 is connected to the air conditioning apparatus 4, via the external interface 13.
  • the control apparatus 1 may be connected to a drive apparatus for reading data stored in a storage medium or the like, via the external interface 13.
  • the control apparatus 1 may acquire the above control program 121, first operation control learning result data 122, second operation control learning result data 123 and conflict resolution learning result data 124, via the drive apparatus.
  • the above control program 121, first operation control learning result data 122, second operation control learning result data 123 and conflict resolution learning result data 124 may be stored in a storage medium.
  • the storage medium is a medium that stores recorded information such as programs and the like through a mechanical, electrical, magnetic, optical or chemical operation, so that the information such as programs and the like is readable by a computer or other apparatus, machine or the like.
  • the storage medium is, for example, a CD (Compact Disc), a DVD (Digital Versatile Disc), a flash memory, or the like.
  • FIG. 3 schematically illustrates an example of the hardware configuration of the data collection control apparatus 2 according to the present embodiment.
  • the data collection control apparatus 2 is a control apparatus that is utilized in situations in which learning data is collected, and is constituted substantially similarly to the above control apparatus 1. That is, the control apparatus 2 is a computer to which a data collection control unit 21, a storage unit 22 and an external interface 23 are electrically connected. Note that, in Fig. 3, the external interface is described as "external I/F", similarly to the above Fig. 2.
  • the control unit 21, the storage unit 22 and the external interface 23 are constituted similarly to the control unit 11, the storage unit 12 and the external interface 13 of the above control apparatus 1.
  • the data collection control program 221 is a program for causing the data collection control apparatus 2 to execute processing for collecting learning data which will be discussed later (Fig. 9).
  • the learning data 223 is data for performing learning of the third learning machine so as to output a control value corrected so as to resolve conflict, when control values that are respectively output from the first learning machine and the second learning machine are input. A detailed description thereof will be given later. Learning Apparatus
  • Fig. 4 schematically illustrates an example of the hardware configuration of the learning apparatus 3 according to the present embodiment.
  • the learning apparatus 3 is a computer to which a control unit 31, a storage unit 32, a communication interface 33, an input apparatus 34, an output apparatus 35 and a drive 36 are electrically connected.
  • the communication interface is described as "communication I/F”.
  • the control unit 31 includes a CPU, a RAM, a ROM and the like, and is configured so as to execute various information processing based on programs and data.
  • the storage unit 32 stores a learning program 321 that is executed with the control unit 31, learning data 223 that is utilized in learning of the third learning machine, 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 apparatus 3 to execute learning processing (Fig. 10) which will be discussed later.
  • the communication interface 33 is an interface for performing wired or wireless communication via a network, such as a cable LAN (Local Area Network) module, a wireless LAN module or the like, for example.
  • the input apparatus 34 is an apparatus for performing input such as a mouse, a keyboard or the like, for example.
  • the output apparatus 35 is an apparatus for performing output such as a display, a speaker or the like, for example.
  • the drive 36 is a drive apparatus for reading programs stored in a storage medium 91, such as a CD drive, a DVD drive or the like, for example.
  • the type of drive 36 may be selected as appropriate according to the type of storage medium 91.
  • the above learning program 321 and/or learning data 223 may be stored in the storage medium 91.
  • the storage medium 91 is a medium that stores recorded information such as programs and the like through a mechanical, electrical, magnetic, optical or chemical operation, so that the information such as programs and the like is readable by a computer or other apparatus, machine or the like.
  • the learning apparatus 3 may acquire the above learning program 321 and/or learning data 223 from this storage medium 91.
  • a disk-type storage medium such as a CD or a DVD is illustrated as an example of the storage medium 91.
  • the type of storage medium 91 need not be limited to a disk-type storage medium, and may be other than a disk-type storage medium.
  • a storage medium other than a disk-type storage medium a semiconductor memory such as a flash memory can be given, for example.
  • Fig. 5 schematically illustrates an example of the functional configuration of the control apparatus 1 according to the present embodiment.
  • the control unit 11 of the control apparatus 1 loads the control program 121 stored in the storage unit 12 to the RAM.
  • the control unit 11 then controls the constituent elements, by interpreting and executing the control program 121 loaded to the RAM using the CPU.
  • the control apparatus 1 thereby functions as a computer that is provided with a first control processing unit 111, a second control processing unit 112 and a conflict resolution unit 113.
  • the first control processing unit 111 controls the operation of the air conditioning apparatus 4, utilizing the first neural network 5 which is the first learning machine.
  • the first neural network 5 has learned control of the operation of the air conditioning apparatus 4 suited to the preference of the user A.
  • the first control processing unit 111 acquires a control value for the air conditioning apparatus 4 from the first neural network 5, by inputting instruction data, position information and the like from the user A to the first neural network 5.
  • the second control processing unit 112 controls the operation of the air conditioning apparatus 4, utilizing the second neural network 6 which is the second learning machine.
  • the second neural network 6 has learned control of the operation of the air conditioning apparatus 4 suited to the preference of the user B.
  • the second control processing unit 112 acquires a control value for the air conditioning apparatus 4 from the second neural network 6, by inputting instruction data, position information and the like from the user B to the second neural network 6.
  • the type of information (data) that is input to the first neural network 5 and second neural network 6 may be determined as appropriate according to the embodiment.
  • the users A and B may, for example, request the air conditioning apparatus 4 to adjust the room temperature, using a user terminal such as a PC (Personal Computer), a mobile phone or a remote controller.
  • the control apparatus 1 may receive instruction data from the user terminals of the users (A, B), through well-known wireless or wired data communication.
  • the control apparatus 1 may acquire various information to be input to the neural networks (5, 6), in accompaniment with the instruction data from the user terminals.
  • the control apparatus 1 may acquire position information of the users (A, B) from the user terminals, as information to be input to the neural networks (5, 6).
  • control apparatus 1 may hold personal information of the users (A, B) in the storage unit 12 in advance.
  • control unit 11 when instruction data is received from the user terminals, may acquire the personal information of the users (A, B) from the storage unit 12, as information to be input to the neural networks (5, 6).
  • the conflict resolution unit 113 in the case where control of the air conditioning apparatus 4 based on the control value that is output from the first neural network 5 conflicts with control of the air conditioning apparatus 4 based on the control value that is output from the second neural network 6, resolves the conflict by correcting control of the air conditioning apparatus 4.
  • the conflict resolution unit 113 resolves the conflict, utilizing the third neural network 7 which is the third learning machine.
  • the third neural network 7 has learned to output a control value corrected so as to resolve conflict, when control values that are respectively output from the first neural network 5 and the second neural network 6 are input.
  • the conflict resolution unit 113 by inputting the control values that are respectively output from the first neural network 5 and the second neural network 6 to the third neural network 7, is able to acquire a control value corrected so as to resolve conflict from the third neural network 7.
  • the first neural network 5 is a neural network having a multilayer structure that is used in so-called deep learning, and is provided with an input layer 51, an intermediate layer (hidden layer) 52 and an output layer 53 in order from input.
  • the first neural network 5 is provided with one intermediate layer 52, with the output of the input layer 51 being the input of the intermediate layer 52, and the output of the intermediate layer 52 being the input of the output layer 53.
  • the number of the intermediate layer 52 need not, however, be limited to one layer, and the first neural network 5 may be provided with two or more intermediate layers 52.
  • Each of the layers 51 to 53 is provided with one or a plurality of neurons.
  • the number of neurons of the input layer 51 can be set according to the number of pieces of information that are utilized in input.
  • the number of neurons of the intermediate layer 52 can be set as appropriate according to the embodiment.
  • the number of neurons of the output layer 53 can be set according to the number of types of control values that are output.
  • each neuron is coupled to all of the neurons in the adjacent layer, but the coupling of neurons need not be limited to such an example, and may be set as appropriate according to the embodiment.
  • a threshold value is set for each neuron, and, basically, the output of each neuron is determined by whether the sum of the products of each input and each weight exceeds the threshold value.
  • the first control processing unit 111 is able to obtain a control value (output value) from the output layer 53, by inputting various information such as instruction data, position information and the like from the user A to the input layer 51 of such a first neural network 5, and determining whether the neurons that are included in the layers 51 to 53 fired in the forward propagation direction.
  • the first control processing unit 111 configures the settings of the first neural network 5 that has learned control of the operation of the air conditioning apparatus 4 suited to the preference of the user A, with reference to the first operation control learning result data 122.
  • the second neural network 6 and third neural network 7 are also constituted similarly to the first neural network 5. That is, the second neural network 6 is provided with an input layer 61, an intermediate layer (hidden layer) 62 and an output layer 63 in order from input.
  • the third neural network 7 is provided with an input layer 71, an intermediate layer (hidden layer) 72 and an output layer 73 in order from input.
  • the number of intermediate layers (62, 72), the number of neurons in each layer (61 to 63, 71 to 73) and the coupling of neurons in adjacent layers may be set as appropriate according to the embodiment.
  • the second control processing unit 112 is able to obtain a control value (output value) from the output layer 63, by inputting various information such as instruction data, position information and the like from the user B to the input layer 61 of the second neural network 6, and determining whether the neurons that are included in the layers 61 to 63 fired in the forward propagation direction.
  • the conflict resolution unit 113 is able to obtain a corrected control value (output value) from the output layer 73, by inputting the control values that are 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 determining whether the neurons that are included in the layers 71 to 73 fired in the forward propagation direction.
  • the second control processing unit 112 configures the settings of the second neural network 6 that has learned control of the operation of the air conditioning apparatus 4 suited to the preference of user B with reference to the second operation control learning result data 123.
  • the conflict resolution unit 113 configures the settings of the third neural network 7 that has learned to output a control value corrected so as to resolve conflict, when control values that are respectively output from the first neural network 5 and the second neural network 6 are input, with reference to the conflict resolution learning result data 124.
  • Fig. 6 schematically illustrates an example of the functional configuration of the data collection control apparatus 2 according to the present embodiment.
  • the control unit 21 of the data collection control apparatus 2 loads the data collection control program 221 stored in the storage unit 22 to the RAM.
  • the control unit 21 then controls the constituent elements, by interpreting and executing the data collection control program 221 loaded to the RAM using the CPU.
  • the data collection control apparatus 2 thereby functions as a computer that is provided with a first control processing unit 211, a second control processing unit 212, a correction value determination unit 213 and a learning data creation unit 214.
  • the first control processing unit 211 is similar to the first control processing unit 111 of the above control apparatus 1. That is, the first control processing unit 211 configures the settings of the first neural network 5, with reference to the first operation control learning result data 122. The first control processing unit 211 then acquires a control value (output value) for the air conditioning apparatus 4 that depends on the preference of the user A from the output layer 53, by inputting various information such as instruction data, position information and the like from the user A to the input layer 51 of the set first neural network 5, and determining whether the neurons that are included in the layers 51 to 53 fired in the forward propagation direction.
  • the second control processing unit 212 is similar to the second control processing unit 112 of the above control apparatus 1. That is, the second control processing unit 212 configures the settings of the second neural network 6, with reference to the second operation control learning result data 123. The second control processing unit 212 then acquires a control value (output value) for the air conditioning apparatus 4 that depends on the preference of the user B from the output layer 63, by inputting various information such as instruction data, position information and the like from the user B to the input layer 61 of the second neural network 6, and determining whether the neurons that are included in the layers 61 to 63 fired in the forward propagation direction.
  • the data collection control apparatus 2 controls the operation of the air conditioning apparatus 4, based on the control values that are respectively obtained from the first neural network 5 and the second neural network 6. In the case where conflict arises in control of the air conditioning apparatus 4, as a result of attempting to control the operation of the air conditioning apparatus 4, the air conditioning apparatus 4 could, however, possibly become inoperative.
  • control value output from the first neural network 5 constitutes a command for raising the room temperature to 26 degrees
  • control value output from the second neural network 6 constitutes a command for lowering the room temperature to 22 degrees, in a situation where the room temperature is 24.
  • conflict will occur in control of the air conditioning apparatus 4, and the data collection control apparatus 2 will become unable to judge whether to control the operation of the air conditioning apparatus 4 to raise the room temperature or to control the operation of the air conditioning apparatus 4 to lower the room temperature.
  • the correction value determination unit 213 determines a correction value of the control values so as to resolve the conflict.
  • the learning data creation unit 214 then creates the learning data 223 for building the third neural network 7, with the control values that are obtained from the neural networks (5, 6) as input data, and with the corrected control value determined by the correction value determination unit 213 as training data.
  • Fig. 7 schematically illustrates an example of the functional configuration of the learning apparatus 3 according to the present embodiment.
  • the control unit 31 of the learning apparatus 3 loads the learning program 321 stored in the storage unit 32 to the RAM.
  • the control unit 31 then controls the constituent elements, by interpreting and executing the learning program 321 loaded to the RAM using the CPU.
  • the learning apparatus 3 thereby functions as a computer that is provided with a learning data acquisition unit 311 and a learning processing unit 312.
  • the learning data acquisition unit 311 acquires the learning data 223 created as described above.
  • the learning processing unit 312 builds the third neural network that is used by the above control apparatus 1, utilizing the acquired learning data 223 and a learning neural network 8. That is, the learning processing unit 312 trains the neural network 8 to output a control value corrected so as to resolve conflict, when control values that are obtained from the neural networks (5, 6) are input.
  • the neural network 8 that serves as a learning target is constituted similarly to the third neural network 7. That is, the learning neural network 8 is provided with an input layer 81, an intermediate layer (hidden layer) 82 and an output layer 83, and the layers 81 to 83 are constituted similarly to the layers 71 to 73 of the above third neural network 7.
  • the learning processing unit 312 builds the neural network 8 that outputs a control value corrected so as to resolve conflict from the output layer 83, when control values that are obtained from the neural networks (5, 6) are input to the input layer 81, through processing for training the neural network.
  • the neural network 8 built in this way is utilizable as the trained third neural network 7.
  • the learning processing unit 312 stores information indicating the configuration of the built neural network 8, the weight of coupling between neurons and the threshold value of each neuron in the storage unit 32 as the conflict resolution learning result data 124.
  • control apparatus 1 The functions of the control apparatus 1, the data collection control apparatus 2 and the learning apparatus 3 will be discussed in detail with operation examples which will be described later. Note that, in the present embodiment, an example will be described in which the functions of the control apparatus 1, the data collection control apparatus 2 and the learning apparatus 3 are all realized by a general-purpose CPU. However, some or all of the above functions may be realized by one or a plurality of dedicated hardware processors. Also, in relation to the respective functional configurations of the control apparatus 1, the data collection control apparatus 2 and the learning apparatus 3, functions may be omitted, replaced or added, according to the embodiment. 3 Operation Examples Control Apparatus
  • Fig. 8 is a flowchart illustrating an example of the processing procedure of the control apparatus 1. Note that the processing procedure that will be described below is merely an example, and the various processing may be changed to the extent possible. Also, with regard to the processing procedure that will be described below, steps can be omitted, replaced or added as appropriate, according to the embodiment. Startup
  • Step S101 a system including the control apparatus 1 and the air conditioning apparatus 4 is started up as appropriate.
  • the control apparatus 1 reads the control program 121 and executes initialization processing. Specifically, the control unit 11 sets the structure of the neural networks 5 to 7, the weight of coupling between neurons and the threshold value of each neuron, with reference to the learning result data 122 to 124. The control unit 11 then controls the operation of the air conditioning apparatus 4, in accordance with the following processing procedure.
  • step S101 the control unit 11 acquires various information that is utilized in control of the operation of the air conditioning apparatus 4, or in other words, information that will be a factor in determining the operation of the air conditioning apparatus 4, from the users (A, B).
  • the type of information that is utilized in control of the operation of the air conditioning apparatus 4 may be determined as appropriate according to the above embodiment.
  • the users (A, B) may request the air conditioning apparatus 4 to adjust the room temperature, using user terminals such as PCs (Personal Computers), mobile phones and remote controllers, for example.
  • the control unit 11 may acquire, from the user terminals, various information such as instruction data, position information and the like that is utilized in control of the operation of the air conditioning apparatus 4, through well-known wireless or wired data communication.
  • the control unit 11 may acquire personal information of the users (A, B) from the storage unit 12, as information to be utilized in control of the operation of the air conditioning apparatus 4.
  • the control unit 11 functions as the first control processing unit 111, and inputs various information such as instruction data, position information and the like acquired from the user A to the first neural network 5.
  • the first neural network 5 has learned to output a control value for the air conditioning apparatus 4 that depends on the preference of the user A, when various information acquired from the user A is input.
  • the control unit 11 is able to acquire, from each neuron of the output layer 53, a control value for the air conditioning apparatus 4 that depends on the preference of the user A, by inputting various information acquired from the user A to each neuron of the input layer 51, and determining whether the neurons that are included in the layers 51 to 53 fired in the forward propagation direction.
  • control unit 11 functions as the second control processing unit 112, and inputs various information such as instruction data, position information and the like acquired from the user B to the second neural network 6.
  • the second neural network 6 has learned to output a control value for the air conditioning apparatus 4 that depends on the preference of the user B, when various information acquired from the user B is input.
  • the control unit 11 is able to acquire, from each neuron of the output layer 63, a control value for the air conditioning apparatus 4 that depends on the preference of the user B, by inputting various information acquired from the user B to each neuron of the input layer 61, and determining whether the neurons that are included in the layers 61 to 63 fired in the forward propagation direction.
  • control unit 11 functions as the conflict resolution unit 113, and inputs the control values obtained from the neural networks (5, 6) to the input layer 71 of the third neural network 7.
  • the control unit 11 acquires a control value corrected so as to resolve conflict from the output layer 73 of the third neural network 7, by determining whether the neurons that are included in the layers 71 to 73 fired in the forward propagation direction.
  • control of the air conditioning apparatus 4 can be corrected so as to resolve the conflict.
  • control values acquired at step S102 are input to the third neural network 7, without distinguishing whether conflict arises in control of the air conditioning apparatus 4 due to these control values.
  • control values acquired at step S102 are also input to the third neural network 7, in the case where conflict does not arise in control of the air conditioning apparatus 4 due to the control values acquired at step S102.
  • the third neural network 7 in the case where conflict does not occur in control of the air conditioning apparatus 4 due to the control values acquired at step S102, may have learned to output the control values directly, or may have learned to output a corrected control value similarly to the case where conflict arises.
  • a control value that is output directly from the third neural network 7 without correcting the input control value will also be referred to as "a control value that has been corrected (corrected control value)".
  • control unit 11 controls the operation of the air conditioning apparatus 4, based on the corrected control value acquired from the third neural network 7 in the above step S103.
  • the control value indicates, for example, the desired room temperature to be attained by operating the air conditioning apparatus 4.
  • the control unit 11 compares the desired room temperature indicated by the control value with the current room temperature, and controls the air conditioning operation of the air conditioning apparatus 4 such that the desired room temperature is attained.
  • control unit 11 ends the processing related to this operation example.
  • the control unit 11 may repeatedly execute the processing of the above steps S101 to S104 regularly or irregularly.
  • the control apparatus 1 is thereby able to continuously implement control the operation of the air conditioning apparatus 4 that depends on the preferences of the users (A, B).
  • Fig. 9 is a flowchart illustrating an example of the processing procedure of the data collection control apparatus 2.
  • the processing procedure that will be described below corresponds to the "learning data creation method" of the present invention.
  • the processing procedure that will be described below is, however, merely an example, and the various processing may be changed to the extent possible. Also, with regard to the processing procedure that will be described below, steps can be omitted, replaced or added as appropriate, according to the embodiment. Startup
  • a system including the data collection control apparatus 2 and the air conditioning apparatus 4 is started up as appropriate, similarly to the above.
  • the data collection control apparatus 2 reads the data collection control program 221 and executes initialization processing. That is, the control unit 21 sets the structure of the neural networks (5, 6), the weight of coupling between neurons and the threshold value of each neuron, with reference to the learning result data (122, 123). The control unit 21 then creates the learning data 223 for building the third neural network 7, in accordance with the following processing procedure. Steps S201 and S202
  • step S201 the control unit 21, similarly to the above step S101, acquires various information to be input to the neural networks (5, 6) from the users (A, B).
  • control unit 21 similarly to the above step S102, functions as the first control processing unit 211, and inputs various information acquired from the user A to the input layer 51 of the first neural network 5.
  • the control unit 21 acquires a control value for the air conditioning apparatus 4 that depends on the preference of the user A that is output from the output layer 53 of the first neural network 5, by determining whether the neurons that are included in the layers 51 to 53 fired in the forward propagation direction.
  • control unit 21 functions as the 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.
  • the control unit 21 then acquires a control value for the air conditioning apparatus 4 that depends on the preference of the user B that is output from the output layer 63 of the second neural network 6, by determining whether the neurons that are included in the layers 61 to 63 fired in the forward propagation direction. Steps S203 and S204
  • step S203 the control unit 21 controls the operation of the air conditioning apparatus 4 which is the control target apparatus, based on the control values acquired from the neural networks (5, 6) at the above step S202.
  • step S204 the control unit 21 then determines whether conflict arises in control of the air conditioning apparatus 4.
  • control unit 21 may determine whether the control values acquired from the neural networks (5, 6) cause conflict to arise, by actually operating the air conditioning apparatus 4 in step S203. Also, the control unit 21 may determine whether conflict arises, by simulating the operation of the air conditioning apparatus 4, based on the control values acquired from the neural networks (5, 6), without actually operating the air conditioning apparatus 4 in step S203.
  • the method of determining whether conflict arises may be set as appropriate according to the embodiment. For example, in the case where control of the air conditioning apparatus 4 that is based on the control values acquired from the neural networks (5, 6) cannot be simultaneously executed, the control unit 21 may determine that conflict arises in control of the air conditioning apparatus 4. In the case where it is determined that conflict arises in control of the air conditioning apparatus 4, the control unit 21 advances the processing to the next step S205. On the other hand, in the case where it is determined that conflict does not arise in control of the air conditioning apparatus 4, the control unit 21 ends the processing related to this operation example. Step S205
  • control unit 21 functions as the correction value determination unit 213, and determines the correction value of the control values acquired from the neural networks (5, 6), so as to resolve conflict that occurs in control of the air conditioning apparatus 4.
  • the control unit 21 thereby acquires a control value corrected so as to resolve conflict.
  • the method of correcting control values may be selected as appropriate according to the embodiment.
  • the control unit 21 may determine the correction value in accordance with a predetermined rule.
  • the control unit 21 prioritizes one of the control values acquired from the neural networks (5, 6). That is, the control unit 21 handles the control value that is prioritized as the corrected control value.
  • the control unit 21 acquires the corrected control value by averaging the control values acquired from the neural networks (5, 6).
  • the control unit 21 may acquire a weighted average of the control values acquired from the neural networks (5, 6) as the corrected control value.
  • control unit 21 may accept input of a correction value from an operator. That is, the control unit 21 may determine the corrected control value based on an input from an operator.
  • the data collection control apparatus 2 may be connected to an input apparatus such as a keyboard, a microphone or the like, via the external interface 23. The operator is able thereby to input a corrected control value by keyboard input, audio input, or the like.
  • control unit 21 functions as the learning data creation unit 214, and pairs the control values acquired from the neural networks (5, 6) at step S202 with the corrected control value determined at step S205.
  • the control unit 21 thereby creates the learning data 223 with the control values before correction as input data, and with the corrected control value as training data.
  • the control unit 21 then saves the created learning data 223 in the storage unit 22.
  • control unit 21 ends the processing related to this operation example.
  • the control unit 21 is able to collect a plurality of pieces of the learning data 223, by repeatedly performing the series of processing of the above steps S201 to S206.
  • control unit 21 may create the learning data 223 with the control values at the time at which it is determined at step S204 that conflict does not arise as input data and training data.
  • control unit 21 may execute the processing of the above steps S205 and S206, even when it is determined at step S204 that conflict does not arise.
  • control unit 21 may accept input from the operator, and determine the correction value such that a corrected control value suited to both of the users A and B is obtained.
  • the control unit 21 is thereby able to create learning data 223 that is utilizable in building a third neural network 7 that outputs a corrected control value that suits the preferences of both the users A and B, when control values that are obtained from the neural networks (5, 6) are input.
  • Fig. 10 is a flowchart illustrating an example of the processing procedure of the learning apparatus 3. Note that the processing procedure that will be described below is merely an example, and the various processing may be changed to the extent possible. Also, with regard to the processing procedure that will be described below, steps can be omitted, replaced or added as appropriate, according to the embodiment. Step S301
  • control unit 31 functions as the learning data acquisition unit 311, and acquires the learning data 223 created by the above data collection control apparatus 2.
  • the method of transferring the learning data 223 created by the data collection control apparatus 2 to the learning apparatus 3 may be selected as appropriate according to the embodiment.
  • the control unit 31 is able to acquire the learning data 223, by accessing the data collection control apparatus 2 via the network.
  • the learning data 223 created with the data collection control apparatus 2 may be stored in another information processing apparatus (storage device) such as a NAS (Network Attached Storage).
  • the control unit 31 is able to acquire the learning data 223, by accessing this other information processing apparatus.
  • the learning data 223 created with the data collection control apparatus 2 may be stored in the storage medium 91.
  • control unit 31 is able to acquire the learning data 223 from the storage medium 91 via the drive 36.
  • the number of pieces of learning data 223 that are acquired in this step S301 may be determined as appropriate according to the embodiment, so as to able to perform learning of the learning neural network 8.
  • control unit 31 functions as the learning processing unit 312, and performs learning of the learning neural network 8 such that a control value corrected so as to resolve conflict is output, when control values that are obtained from the neural networks (5, 6) are input, using the learning data 223 acquired at step S301.
  • the control unit 31 prepares the learning neural network 8 that serves as a target for performing learning processing.
  • the configuration of the prepared neural network 8, the initial value of the weight of coupling between neurons and the initial value of the threshold value of each neuron may be supplied by a template or may be supplied by an input from the operator.
  • the control unit 31 may prepare the learning neural network 8, based on the conflict resolution learning result data 124 that serves as a target for performing relearning.
  • control unit 31 performs learning of the neural network 8, with the control values obtained from the neural networks (5, 6) that are included in the learning data 223 acquired at step S301 as input data, and with the corrected control value as training data.
  • Methods such as gradient descent or stochastic gradient descent may be used in this learning of the neural network 8.
  • control unit 31 inputs the control values obtained from the neural networks (5, 6) that are included in the learning data 223 to the input layer 81, and performs computational processing of the learning neural network 8 in the forward propagation direction.
  • the control unit 31 thereby obtains an output value from the output layer 83 of the learning neural network 8.
  • the control unit 31 calculates the error of the output value that is output from the output layer 83 with the corrected control value included in the learning data 223.
  • the control unit 31 calculates the respective errors of the weight of coupling between neurons and the threshold value of each neuron, through error back propagation, using the calculated error of the output value.
  • the control unit 31 updates the respective threshold values of the weight of coupling between neurons and the threshold value of each neuron, based on the calculated errors.
  • a neural network 8 that outputs a control value corrected so as to resolve conflict can thereby be built, when control values that are obtained from the neural networks (5, 6) are input.
  • control unit 31 functions as the learning processing unit 312, and stores information indicating the configuration of the built neural network 8, the weight of coupling between neurons and the threshold value of each neuron in the storage unit 32 as the conflict resolution learning result data 124.
  • the control unit 31 thereby ends the learning processing related to this operation example.
  • the control unit 31 may transfer the created conflict resolution learning result data 124 to the control apparatus 1 as appropriate, after the processing of the above step S303 is completed. Also, the control unit 31 may regularly or irregularly update the conflict resolution learning result data 124, by regularly or irregularly executing the learning processing of the above steps S301 to S303. The control unit 31 may then regularly or irregularly update the conflict resolution learning result data 124 that is held by the control apparatus 1, by transferring the created conflict resolution learning result data 124 to the control apparatus 1 whenever the learning processing is executed. Operation and Effect
  • the operation of the air conditioning apparatus 4 can be controlled to suit the preferences of the users (A, B), by the neural networks (5, 6). In cases such as where contradictory instructions are given by both users, conflict can arise in control of the air conditioning apparatus 4.
  • the control values obtained from the neural networks (5, 6) can be corrected so as to resolve the conflict, by a third neural network 7 that is utilized in the processing of the above step S103. Accordingly, with the present embodiment, it is possible to ensure that, even when conflict arises between controls by the users (A, B), the air conditioning apparatus 4 does not become inoperative. 4 Variations
  • an air conditioning apparatus is illustrated as a control target apparatus that is controlled by the control apparatus 1.
  • the type of control target apparatus need not be limited to an air conditioning apparatus, and may be selected as appropriate according to the embodiment.
  • the control target apparatus may be a robotic apparatus or the like, for example.
  • the first neural network 5 and second neural network 6 take the same control target apparatus (air conditioning apparatus 4) as the target for control.
  • the control target apparatus that the first neural network 5 takes as the target for control may be different from the control target apparatus that the second neural network 6 takes as the target for control.
  • the first neural network 5 may take, as a first control target apparatus, a first robotic apparatus as the target for control.
  • the second neural network 6 may take, as the second control target apparatus, a second robotic apparatus that is different from the first robotic apparatus as the target for control.
  • conflict can arise between the controls of both robotic apparatuses.
  • two learning machines (first neural network 5 and second neural network 6) are utilized as learning machines that issue control values for controlling the operation of the control target apparatus.
  • the number of learning machines that issue control values for controlling the operation of the control target apparatus need not be limited to two, and may be three or more.
  • control unit 11 may include a plurality of processors.
  • the control apparatus 1 and the data collection control apparatus 2 are provided with a communication interface, and are configured to be capable of exchanging data with another information processing apparatus via a network.
  • the control apparatus 1, the data collection control apparatus 2 and the learning apparatus 3 may respectively be constituted by a plurality of computers.
  • control apparatus 1 and the data collection control apparatus 2 a general-purpose desktop PC, a tablet PC, a mobile phone or the like may be used as appropriate, according to the control target apparatus that serves as a target to be controlled, in addition to an information processing apparatus such as an ECU (Electronic Control Unit) designed exclusively for services that are provided or the like.
  • control target apparatus such as an ECU (Electronic Control Unit) designed exclusively for services that are provided or the like.
  • learning apparatus 3 a general-purpose server apparatus, a desktop PC or the like may be used, in addition to an information processing apparatus designed exclusively for services that are provided.
  • neural networks 5 to 8 general forward propagation neural networks having a multilayer structure are used as the neural networks 5 to 8.
  • the type of the neural networks 5 to 8 need not be limited to such an example, and may be selected as appropriate according to the embodiment.
  • convolutional neural networks provided with a convolutional layer and a pooling layer may be used for the neural networks 5 to 8.
  • time series data for example, recursive neural networks having a coupling that recurses from the output side to the input side such as from the intermediate layer to the input layer may be used for the neural networks 5 to 8.
  • the number of layers, the number of neurons in each layer, the coupling relationship of neurons and the transfer function of each neuron of the neural networks 5 to 8 may be determined as appropriate according to the embodiment.
  • control values acquired from the neural networks (5, 6) at step S102 are input to the third neural network 7, without distinguishing whether the control values cause conflict to arise in control of the air conditioning apparatus 4.
  • processing procedure of the control apparatus 1 need not be limited to such an example.
  • the control unit 11 may input the control values acquired at step S102 to the third neural network 7, only in the case where the control values acquired from the neural networks (5, 6) at step S102 cause conflict to arise in control of the air conditioning apparatus 4.
  • control unit 11 may utilize the control values acquired from the neural networks (5, 6) directly, by omitting the above step S103 and executing the processing of the next step S104.
  • step S103 resolution of conflict that occurs in control of the air conditioning apparatus 4 is performed using the third neural network 7.
  • the method of resolving conflict that occurs in control of the air conditioning apparatus 4 need not be limited to such an example. Resolution of conflict that occurs in control of the air conditioning apparatus 4 may be performed, without using neural networks.
  • Fig. 11 schematically illustrates a control apparatus 1A according to this variation.
  • the control apparatus 1A is constituted similarly to the above control apparatus 1, except for being provided with a conflict resolution unit 113A that does not hold conflict resolution learning result data 124 or utilize a neural network.
  • the control unit 11 in the above step S103, functions as the conflict resolution unit 113A, and determines whether conflict arises in control of the air conditioning apparatus 4 due to the control values acquired from the neural networks (5, 6). For example, the control unit 11 determines whether conflict arises in control of the air conditioning apparatus 4, by simulating the operation of the air conditioning apparatus 4.
  • control unit 11 controls the operation of the air conditioning apparatus 4 based on the control values acquired from the neural networks (5, 6).
  • the control unit 11 corrects the control values acquired from the neural networks (5, 6) so as to resolve the conflict, and controls the operation of the air conditioning apparatus 4 based on the corrected control value.
  • control unit 11 may determine a corrected correction value, by prioritizing one of the control values acquired from the neural networks (5, 6). Also, for example, the control unit 11 may determine a corrected control value, by averaging the control values acquired from the neural networks (5, 6).
  • control unit 11 inputs the control values acquired from the neural networks (5, 6) to the third neural network 7, without specifying what kind of conflict the control values cause in control of the air conditioning apparatus 4.
  • processing procedure of the control apparatus 1 need not be limited to such an example, and the control unit 11 may specify what kind of conflict the control values acquired from the neural networks (5, 6) cause in control of the air conditioning apparatus 4.
  • Fig. 12 schematically illustrates a control apparatus 1B according to this variation.
  • the control apparatus 1B is constituted similarly to the above control apparatus 1, except for being provided with a conflict type specification unit 114 that specifies conflict type information 125 indicating how controls of the air conditioning apparatus 4 conflict, based on the control values acquired from the neural networks (5, 6), and for utilizing the specified conflict type information 125 as input of the third neural network 7.
  • the control unit 11 specifies the conflict type information 125 indicating how controls of the air conditioning apparatus 4 conflict, based on the control values acquired from the neural networks (5, 6), before executing the above step S103.
  • the type (manner) of conflict may be set as appropriate according to the embodiment.
  • the control unit 11 then inputs the control values acquired from the neural networks (5, 6) and the conflict type information 125 to the input layer 71 of the third neural network 7.
  • the control apparatus 1B is thereby able to reliably change the method of correcting control values according to the type of conflict.
  • the third neural network 7 may be configured to output, as the corrected control value, an average value of the control values acquired from the neural networks (5, 6).
  • the third neural network 7 may be configured to preferentially output, as the corrected control value, the control value acquired from the neural network 6.
  • FIG. 13 schematically illustrates a control apparatus 1C according to this variation.
  • the control apparatus 1C is constituted similarly to the above control apparatus 1B, except for being provided with a conflict type specification unit 114C that specifies the conflict type information 125 utilizing a fourth neural network 115.
  • the fourth neural network 115 has learned to output an output value corresponding to the conflict type information 125, when control values that are output from the neural networks (5, 6) are input.
  • the fourth neural network 115 may, for example, be constituted similarly to the neural networks 5 to 7.
  • control unit 11 sets the structure of the fourth neural network 115, the weight of coupling between neurons and the threshold value of each neuron, with reference to the learning result data 126.
  • Conflict that occurs in control of the air conditioning apparatus 4 can thereby be intricately classified, enabling an appropriate resolution method to be employed for each classification.
  • each learning machine is constituted by a neural network.
  • the type of the learning machines need not be limited to a neural network, and may be selected as appropriate according to the embodiment.
  • a support vector machine, a self-organizing map, a learning machine that performs learning by reinforcement learning or the like may be used, for example.
  • a learning apparatus for creating the above neural networks (5, 6) may be prepared.
  • trained neural networks (5, 6) can be created, using the above learning apparatus 3, by changing the learning data that is utilized in machine learning to learning data for learning control suited to the users (A, B) from the above learning data 223.
  • Learning data for learning controls suited to the users (A, B) can be created by combining various information acquired from each user (A, B) that serves as input data and the original control value suited to the preference of the user (A, B) that serves as training data.
  • the learning apparatus is able to build the trained neural networks (5, 6), and create operation control learning result data (122, 123), by executing the processing of the above steps S301 to 303, utilizing such learning data.
  • a learning apparatus for creating the above fourth neural network 115 may be prepared.
  • a trained fourth learned neural network 115 can be created, using the above learning apparatus 3, by changing learning data that is utilized in machine learning to learning data for learning to specify the type of conflict from the above learning data 223.
  • the learning data for learning to specify the type of conflict can be created by combining the control value acquired from each neural network (5, 6) that serves as input data and an output value corresponding to the conflict type information 125 that serves as training data.
  • the learning apparatus is able to build the trained fourth neural network 115, and create the learning result data 126, by executing the processing of the above steps S301 to 303, utilizing such learning data.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Feedback Control In General (AREA)
  • Air Conditioning Control Device (AREA)

Abstract

L'invention concerne une technologie pour garantir que, même lorsqu'un conflit survient entre une pluralité de commandes, des appareils cibles de commande ne deviennent pas inopérants. Un appareil de commande selon un aspect de la présente invention comprend une première unité de traitement de commande configurée pour commander le fonctionnement d'un premier appareil cible de commande, sur la base d'une valeur de commande émise par une première machine d'apprentissage entraînée qui a effectué un apprentissage pour commander le fonctionnement du premier appareil cible de commande, une seconde unité de traitement de commande configurée pour commander le fonctionnement d'un second appareil cible de commande, sur la base d'une valeur de commande émise par une seconde machine d'apprentissage entraînée qui a effectué un apprentissage pour commander le fonctionnement du second appareil cible de commande, et une unité de résolution de conflit configurée pour, lorsque la commande du premier appareil cible de commande sur la base de la valeur de commande émise par la première machine d'apprentissage est en conflit avec la commande du second appareil cible de commande sur la base de la valeur de commande émise par la seconde machine d'apprentissage, résoudre le conflit par correction de la commande des premier et second appareils cible de commande.
PCT/JP2018/016703 2017-05-15 2018-04-25 Appareil de commande, programme de commande, procédé de création de données d'apprentissage et procédé d'apprentissage WO2018211927A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2017096165A JP6915371B2 (ja) 2017-05-15 2017-05-15 制御装置、制御プログラム、学習データ作成方法、及び学習方法
JP2017-096165 2017-05-15

Publications (1)

Publication Number Publication Date
WO2018211927A1 true WO2018211927A1 (fr) 2018-11-22

Family

ID=62200496

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2018/016703 WO2018211927A1 (fr) 2017-05-15 2018-04-25 Appareil de commande, programme de commande, procédé de création de données d'apprentissage et procédé d'apprentissage

Country Status (2)

Country Link
JP (1) JP6915371B2 (fr)
WO (1) WO2018211927A1 (fr)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114675539A (zh) * 2022-03-29 2022-06-28 江苏希尔登家居有限公司 一种自主学习的门窗智能控制系统

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP7121500B2 (ja) * 2018-02-21 2022-08-18 清水建設株式会社 装置制御システム及び装置制御方法
JP2020063343A (ja) 2018-10-16 2020-04-23 日東シンコー株式会社 樹脂組成物

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6594684B1 (en) * 1998-06-15 2003-07-15 Dejima, Inc. Adaptive interaction using an adaptive agent-oriented software architecture
WO2004018158A2 (fr) * 2002-08-21 2004-03-04 Neal Solomon Systemes, procedes et appareils pour l'organisation de groupes d'agents robotiques mobiles auto-configurables dans un systeme multirobotique
JP2016532355A (ja) 2013-07-12 2016-10-13 李 文 嵩 インテリジェント住宅システム及び操作方法

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6594684B1 (en) * 1998-06-15 2003-07-15 Dejima, Inc. Adaptive interaction using an adaptive agent-oriented software architecture
WO2004018158A2 (fr) * 2002-08-21 2004-03-04 Neal Solomon Systemes, procedes et appareils pour l'organisation de groupes d'agents robotiques mobiles auto-configurables dans un systeme multirobotique
JP2016532355A (ja) 2013-07-12 2016-10-13 李 文 嵩 インテリジェント住宅システム及び操作方法

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
ANA L C BAZZAN: "Opportunities for multiagent systems and multiagent reinforcement learning in traffic control", AUTONOMOUS AGENTS AND MULTI-AGENT SYSTEMS, KLUWER ACADEMIC PUBLISHERS, BO, vol. 18, no. 3, 7 September 2008 (2008-09-07), pages 342 - 375, XP019668772, ISSN: 1573-7454 *
FERNANDO LOPES ET AL: "Negotiation among autonomous computational agents: principles, analysis and challenges", ARTIFICIAL INTELLIGENCE REVIEW, KLUWER ACADEMIC PUBLISHERS, DO, vol. 29, no. 1, 15 July 2009 (2009-07-15), pages 1 - 44, XP019727185, ISSN: 1573-7462 *
JUNG K ED - CALDERARA SIMONE ET AL: "Neural network-based text location in color images", PATTERN RECOGNITION LETT, ELSEVIER, AMSTERDAM, NL, vol. 22, no. 14, 1 December 2001 (2001-12-01), pages 1503 - 1515, XP004303677, ISSN: 0167-8655, DOI: 10.1016/S0167-8655(01)00096-4 *
LEE ET AL: "Conflict resolution in multi-agent based Intelligent Environments", BUILDING AND ENVIRONMENT, PERGAMON PRESS, OXFORD, GB, vol. 45, no. 3, 1 March 2010 (2010-03-01), pages 574 - 585, XP026740480, ISSN: 0360-1323, [retrieved on 20090726], DOI: 10.1016/J.BUILDENV.2009.07.013 *
YI JIAO ET AL: "Analyses of conflict among exception in multi-agent system and an integrated management mechanism", COMPUTER SUPPORTED COOPERATIVE WORK IN DESIGN, 2005. PROCEEDINGS OF TH E NINTH INTERNATIONAL CONFERENCE ON COVENTRY, UK MAY 24-26, 2005, PISCATAWAY, NJ, USA,IEEE, vol. 1, 24 May 2005 (2005-05-24), pages 429 - 433, XP010832926, ISBN: 978-1-84600-002-7 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114675539A (zh) * 2022-03-29 2022-06-28 江苏希尔登家居有限公司 一种自主学习的门窗智能控制系统
CN114675539B (zh) * 2022-03-29 2024-01-23 江苏希尔登家居有限公司 一种自主学习的门窗智能控制系统

Also Published As

Publication number Publication date
JP2018194932A (ja) 2018-12-06
JP6915371B2 (ja) 2021-08-04

Similar Documents

Publication Publication Date Title
WO2018211927A1 (fr) Appareil de commande, programme de commande, procédé de création de données d'apprentissage et procédé d'apprentissage
JP2020038699A (ja) 汎用学習済モデルの生成方法
US20120298763A1 (en) Distributed networked thermostat system and method
JP7003480B2 (ja) 自動学習ユニバーサルゲートウェイ
US20190064759A1 (en) Method for creating prototype and apparatus therefor
KR20190043019A (ko) 외부 전자 장치의 상태에 대응하도록 IoT 장치를 제어할 수 있는 전자 장치 및 전자 장치 운용 방법
EP3554043A1 (fr) Procédé et appareil de fourniture de dispositifs sécurisés dans un système de commande industriel
CN114928538A (zh) 设备配置处理方法、装置、计算机设备和存储介质
WO2016170618A1 (fr) Dispositif d'essai automatique
CN108063701B (zh) 一种控制智能设备的方法及装置
JP6298524B2 (ja) デバイスのマッチングのための方法、装置およびシステム
US20220207289A1 (en) Device selection apparatus, data set selection apparatus, method for selecting device, and program
KR102301291B1 (ko) 프로토타입 제작 방법 및 그 장치
US10218722B2 (en) Method and system for developing an anomaly detector for detecting an anomaly parameter on network terminals in a distributed network
CN109782616B (zh) 基于感应阵列的控制方法、装置、存储介质及计算机设备
KR20220083221A (ko) IoT 환경의 허브 장치 및 로컬 네트워크 기반 이벤트 처리 방법
WO2014102836A1 (fr) Procédé de configuration automatique de terminaux portatifs
CN110999263B (zh) Iot装置集群的分层数据处理
US20230254178A1 (en) Server for controlling home network based on sleep state and method for operating the same
WO2020105161A1 (fr) Système de commutation de modèle d'apprentissage automatique de dispositif périphérique, procédé de commutation de modèle d'apprentissage automatique de dispositif périphérique, programme et dispositif périphérique
EP3796239B1 (fr) Procédé et dispositif pour générer un projet d'automatisation de bâtiment
JP7437717B2 (ja) 管理システム、管理方法及びプログラム
US11775403B2 (en) Method and system for developing an anomaly detector for detecting an anomaly parameter on network terminals in a distributed network
US20190182792A1 (en) Network connection control method by using a user context
JP7429878B2 (ja) 位置推定システム、固定無線装置、および移動無線装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18725931

Country of ref document: EP

Kind code of ref document: A1

DPE1 Request for preliminary examination filed after expiration of 19th month from priority date (pct application filed from 20040101)
NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18725931

Country of ref document: EP

Kind code of ref document: A1

点击 这是indexloc提供的php浏览器服务，不要输入任何密码和下载