JP7309069B2 - Learning device and reasoning device for control of air conditioner - Google Patents

Description

TECHNICAL FIELD The present disclosure relates to a learning device and a reasoning device for controlling an air conditioner.

Methods are known for optimal operation control of air conditioning systems. For example, Patent Literature 1 describes a method of controlling the operation of an air conditioner by determining an energy consumption function using measurement data during operation of the air conditioning system.

JP 2006-207929 A

However, in the method for controlling the operation of an air conditioner disclosed in Patent Document 1, a temperature sensor that detects the temperature in the room is fixed and arranged at a predetermined location in the room. Therefore, it may not be possible to set the temperature of the location where the user is located to the set temperature. In particular, such a problem occurs when air currents are disturbed by indoor fixtures and the like.

SUMMARY OF THE INVENTION Therefore, an object of the present disclosure is to provide a learning device and a reasoning device for operation control of an air conditioner that can set the temperature of the location where the user is located to the set temperature.

The learning device for an air conditioner of the present disclosure includes a user position of the air conditioner, a state including the difference between the detected temperature at the user's position and the set temperature of the indoor unit of the air conditioner, and the indoor unit in the state. a data acquisition unit that acquires learning data including the set air volume and the set wind direction; and a model generation unit that generates a learned model for inferring the set air volume of the indoor unit and the set wind direction of the indoor unit from the difference from the set temperature.

A reasoning device for an air conditioner of the present disclosure includes a data acquisition unit that acquires a position of a user of the air conditioner and a state including a difference between a temperature at the user's position and a set temperature of an indoor unit of the air conditioner. , the position of the user of the air conditioner, and a trained model for inferring the set air volume of the indoor unit and the set wind direction of the indoor unit from the difference between the detected temperature at the user's position and the set temperature of the indoor unit of the air conditioner. and an inference unit that infers the set air volume and the set air direction of the indoor unit from the state acquired by the data acquisition unit.

According to the present disclosure, the temperature of the location where the user is located can be set to the set temperature.

BRIEF DESCRIPTION OF THE DRAWINGS It is a figure showing the structure of the air conditioning system of embodiment. 3 is a diagram representing data input to or output from learning device 10 and inference device 30 of Embodiment 1. FIG. FIG. 4 is a diagram representing an example of user positions; 1 is a diagram showing a configuration of learning device 10 according to Embodiment 1. FIG. 4 is a flow chart relating to learning processing of the learning device 10. FIG. 3 is a diagram showing the configuration of an inference device 30; FIG. 4 is a flowchart showing a procedure for inferring the set air volume and the set wind direction of the indoor unit by the inference device 30. FIG. FIG. 10 is a diagram representing data input to or output from learning device 10 and inference device 30 according to a second embodiment; FIG. 13 is a diagram showing the configuration of learning device 10 of Embodiment 3; (a) to (j) are diagrams showing examples of data obtained from the portable sensor 3 and the control device 2. FIG. FIG. 10(a) to FIG. 10(j) are diagrams showing the positions of the users. (a) to (c) are diagrams for explaining a method of increasing data. (a) to (c) are diagrams for explaining a method of increasing data. 3 is a diagram showing a hardware configuration of a learning device 10, an inference device 30, or a control device 2; FIG.

Embodiments will be described below with reference to the drawings.
Embodiment 1.
FIG. 1 is a diagram showing the configuration of an air conditioning system according to an embodiment.

The air conditioning system includes an air conditioner 1 , a control device 2 , a portable sensor 3 , a learning device 10 , a learned model storage section 20 and an inference device 30 .

The portable sensor 3 can be carried by a user. The portable sensor 3 can detect temperature. The portable sensor 3 can detect the user's position and the temperature at the user's position.

The learning device 10 generates a trained model that infers the set air volume and direction of the indoor unit from the user's position and the difference between the detected temperature at the user's position and the set temperature of the indoor unit.

The trained model storage unit 20 stores trained models generated by the learning device 10 .

The inference device 30 appropriately determines the location of the user from the user's position and the difference between the detected temperature at the user's position and the set temperature of the indoor unit according to the learned model stored in the learned model storage unit. In order to achieve the set temperature, the set air volume and the set wind direction of the indoor unit of the air conditioner are estimated.

The control device 2 controls the air conditioner 1 based on the inference result of the inference device 30 and the like.
FIG. 2 is a diagram representing data input to or output from learning apparatus 10 and inference apparatus 30 according to the first embodiment.

B1 (behavior) is the set air volume of the indoor unit and the set wind direction of the indoor unit. B2 (state) is the user's position and the difference between the detected temperature at the user's position and the set temperature of the indoor unit. C (output) is the set air volume of the indoor unit and the set air direction of the indoor unit. D (reward criterion) is the amount of change in the detected temperature of the user's location per unit time.

FIG. 3 is a diagram representing an example of user positions.
By carrying the portable sensor 3 by the user, the position of the user can be detected by the portable sensor 3 . By sensing the temperature of the user's location, it is possible to control the airflow at the user's location.

FIG. 4 is a diagram showing the configuration of learning device 10 according to the first embodiment. The learning device 10 includes a data acquisition section 12 and a model generation section 13 .

The data acquisition unit 12 acquires learning data including B1 (behavior) and B2 (state). That is, the data acquisition unit 12 acquires learning data including the set air volume of the indoor unit, the set air direction of the indoor unit, the position of the user, and the difference between the detected temperature at the user's position and the set temperature of the indoor unit.

The model generation unit 13 generates a trained model that infers C (output) from B2 (state) using learning data including B1 (behavior) and B2 (state) acquired by the data acquisition unit 12 . That is, the model generation unit 13 uses learning data including the set air volume of the indoor unit, the set air direction of the indoor unit, the position of the user, and the difference between the detected temperature at the user's position and the set temperature of the indoor unit. , the position of the user, and a trained model that infers the set air volume and the wind direction setting of the indoor unit from the difference between the detected temperature at the user's position and the set temperature of the indoor unit. The model generation unit 13 stores the generated learned model in the learned model storage unit 20 .

A known algorithm such as supervised learning, unsupervised learning, or reinforcement learning can be used as the learning algorithm used by the model generation unit 13 . As an example, a case where reinforcement learning is applied will be described. In reinforcement learning, an agent (actor) in an environment observes the current state (environmental parameters) and decides what action to take. The environment dynamically changes according to the actions of the agent, and the agent is rewarded according to the change in the environment. The agent repeats this and learns the course of action that yields the most rewards through a series of actions. Q-learning, which is a representative method of reinforcement learning, or TD-learning (Temporal Difference Learning) can be used. For example, in the case of Q-learning, a general update formula for the action-value function Q(s, a) is represented by formula (1).

In equation (1), st represents the state of the environment at time t, and at represents action at time t. Action at causes the state to change to st+1. rt+1 represents the reward obtained by changing the state, γ represents the discount rate, and α represents the learning coefficient. γ is in the range of 0<γ≦1, and α is in the range of 0<α≦1. B1 (action) becomes action at, and B2 (state) becomes state st. That is, the set air volume of the indoor unit and the set air direction of the indoor unit are the action at, and the user's position and the difference between the detected temperature of the user's position and the set temperature of the indoor unit are the state st. In Q-learning, the best action at in state st at time t is learned.

The update formula represented by formula (1) increases the action value Q if the action value Q of action a with the highest Q value at time t+1 is greater than the action value Q of action a executed at time t. On the contrary, the action value Q is decreased. In other words, the action value function Q(s, a) is updated so that the action value Q of action a at time t approaches the best action value at time t+1. As a result, the best behavioral value in a certain environment will be propagated to the behavioral value in the previous environment.

As described above, when generating a trained model by reinforcement learning, the model generator 13 includes the reward calculator 14 and the function updater 15 .

The reward calculator 14 calculates a reward based on B1 (behavior) and B2 (state). That is, the reward calculation unit 14 calculates a reward based on the set air volume of the indoor unit, the set wind direction of the indoor unit, the position of the user, and the difference between the detected temperature at the user's position and the set temperature of the indoor unit. . The reward calculator 14 calculates a reward r based on the amount of temperature change per unit time at the user's position. For example, if the amount of temperature change per unit time at the user's position increases, the reward calculation unit 14 increases the reward r (for example, gives a reward of "1"). If the amount of temperature change per win is reduced, the reward r is reduced (for example, a reward of "-1" is given).

The function updating unit 15 updates the function for determining the set air volume of the indoor unit and the wind direction setting of the indoor unit according to the reward calculated by the reward calculating unit 14 , and outputs the function to the learned model storage unit 20 . For example, in the case of Q-learning, the function updating unit 15 uses the action value function Q(st, at) represented by Equation (1) as a function for calculating the set air volume of the indoor unit and the wind direction setting of the indoor unit. .

The above learning is repeatedly executed. The learned model storage unit 20 stores the action value function Q(st, at) updated by the function update unit 15, that is, the learned model.

FIG. 5 is a flowchart relating to the learning process of the learning device 10. As shown in FIG.
In step S101, the data acquisition unit 12 obtains the set air volume of the indoor unit, the set air direction of the indoor unit, the position of the user, and the difference between the detected temperature at the user's position and the set temperature of the indoor unit. Get training data containing

In step S102, the model generation unit 13 calculates a reward based on the set air volume of the indoor unit, the set air direction of the indoor unit, the position of the user, and the difference between the detected temperature at the user's position and the set temperature of the indoor unit. calculate. Specifically, the reward calculator 14 determines whether to increase or decrease the reward based on the amount of temperature change per unit time at the user's position.

When the remuneration calculation unit 14 determines to increase the remuneration, the process proceeds to step S103. When the remuneration calculation unit 14 determines to decrease the remuneration, the process proceeds to step S104.

In step S103, the reward calculator 14 increases the reward.
In step S104, the reward calculator 14 reduces the reward.

In step S105, the function updating unit 15 updates the action value function Q(st, at) expressed by Equation (1) stored in the trained model storage unit 20 based on the reward calculated by the reward calculation unit 14. Update.

The learning device 10 repeatedly executes steps S101 to S105 described above, and stores the generated action-value function Q(st, at) as a learned model.

Although the learning device 10 according to the present embodiment stores the learned model in the learned model storage unit 20 provided outside the learning device 10, the learned model storage unit 20 is stored inside the learning device 10. be prepared for

FIG. 6 is a diagram showing the configuration of the inference device 30. As shown in FIG. The inference device 30 includes a data acquisition unit 31 and an inference unit 32 .

The data acquisition unit 31 acquires the B2 input. That is, the data acquisition unit 31 acquires the user's position and the difference between the detected temperature at the user's position and the set temperature of the indoor unit.

The inference unit 32 is used to infer the set air volume of the indoor unit and the wind direction setting of the indoor unit from the user's position and the difference between the detected temperature at the user's position and the set temperature of the indoor unit from the learned model storage unit 20. Read the trained model.

The inference unit 32 infers the C output using the data acquired by the data acquisition unit 31 and the learned model. That is, the inference unit 32 inputs the location of the user and the difference between the detected temperature of the user's location and the set temperature of the indoor unit, which are acquired by the data acquisition unit 31, into the trained model, thereby determining the location where the user is located. It is possible to infer the set air volume of the indoor unit and the set wind direction of the indoor unit so that the temperature of the indoor unit becomes the set temperature.

For example, the inference unit 32 reads the action-value function Q(st, at) from the learned model storage unit 20 as a learned model. The inference unit 32 determines the setting of the indoor unit based on the action value function Q(s, a) with respect to the user's position and the difference between the detected temperature at the user's position and the set temperature of the indoor unit (state st). Obtain the wind volume and set wind direction (behavior at) of the indoor unit.

In the present embodiment, the learned model learned by the model generation unit of the air conditioner is used to output the set air volume of the indoor unit and the wind direction setting of the indoor unit. A model may be acquired, and the set air volume of the indoor unit and the wind direction setting of the indoor unit may be output based on this learned model.

FIG. 7 is a flowchart showing a procedure for inferring the set air volume of the indoor unit and the set wind direction of the indoor unit by the inference device 30 .

In step S201, the data acquisition unit 31 acquires the position of the user and the difference between the detected temperature at the position of the user and the set temperature of the indoor unit.

In step S<b>202 , the inference unit 32 inputs the user's position and the difference between the detected temperature at the user's position and the set temperature of the indoor unit to the learned model stored in the learned model storage unit 20 .

In step S203, the inference unit 32 obtains the set air volume of the indoor unit and the set air direction of the indoor unit from the learned model. The inference unit 32 outputs the obtained set air volume of the indoor unit and set air direction of the indoor unit to the control device 2 .

In step S204, the control device 2 uses the output set air volume of the indoor unit and set air direction of the indoor unit. It controls the air conditioner 1 .

In the present embodiment, the case where reinforcement learning is applied to the learning algorithm used by the inference unit has been described, but the present invention is not limited to this. As for the learning algorithm, supervised learning, unsupervised learning, or semi-supervised learning can be applied in addition to reinforcement learning.

As a learning algorithm used in the model generating unit 13, deep learning that learns to extract the feature amount itself can also be used. Alternatively, machine learning may alternatively be performed according to other known methods, such as neural networks, genetic programming, functional logic programming, or support vector machines.

For example, the learning device 10 and the inference device 30 may be connected to the control device 2 via a network and may be separate devices from the control device 2 . Also, the learning device 10 and the reasoning device 30 may be built in the control device 2 . Furthermore, the learning device 10 and the reasoning device 30 may reside on a cloud server.

The model generation unit 13 may learn the set air volume of the indoor unit and the wind direction setting of the indoor unit using learning data acquired from a plurality of air conditioners. Note that the model generating unit 13 may acquire learning data from a plurality of air conditioners used in the same area, or may acquire learning data collected from a plurality of air conditioners operating independently in different areas. The set air volume of the indoor unit and the wind direction setting of the indoor unit may be learned using the data for the indoor unit. Also, it is possible to add or remove an air conditioner from which data for learning is collected on the way. Furthermore, a learning device that has learned the set air volume of an indoor unit and the wind direction setting of an indoor unit for a certain air conditioner is applied to another air conditioner, and the set air volume and air direction of the indoor unit are applied to the other air conditioner. The wind direction setting of the indoor unit may be re-learned and updated.

As described above, according to the present embodiment, the learning device sets the set air volume and the air direction of the indoor unit based on the position of the user and the difference between the detected temperature at the user's position and the set temperature of the indoor unit. A learned model for inference is generated, and the inference device adjusts the location of the user to the set temperature appropriately based on the user's location and the difference between the detected temperature at the user's location and the set temperature of the indoor unit according to the learned model. In order to do so, it is possible to estimate the set air volume and the set wind direction of the indoor unit of the air conditioner.

Embodiment 2.
This embodiment relates to a remuneration standard different from that of the first embodiment.

FIG. 8 is a diagram representing data input to or output from learning apparatus 10 and inference apparatus 30 according to the second embodiment.

B1 (behavior) is the set air volume of the indoor unit and the set wind direction of the indoor unit. B2 (state) is the user's position and the difference between the detected temperature at the user's position and the set temperature of the indoor unit. C (output) is the set air volume of the indoor unit and the set air direction of the indoor unit. D (reward standard) is an operation for setting the wind volume or wind direction by the user.

The reward calculator 14 calculates a reward based on B1 (behavior) and B2 (state). That is, the reward calculation unit 14 calculates a reward based on the set air volume of the indoor unit, the set wind direction of the indoor unit, the position of the user, and the difference between the detected temperature at the user's position and the set temperature of the indoor unit. . The remuneration calculation unit 14 calculates a remuneration r based on the user's operation for setting the wind volume or wind direction. For example, if the user does not set the wind volume or wind direction, the reward calculation unit 14 increases the reward r (for example, gives a reward of “1”). If the operation is executed, the reward r is reduced (for example, a reward of "-1" is given).

Embodiment 3.
FIG. 9 is a diagram showing the configuration of the learning device according to the third embodiment.

The learning device 10 according to the third embodiment differs from the learning device 10 according to the first embodiment in that the learning device 10 according to the third embodiment includes a data extender 62 .

The data extension unit 62 determines the unacquired positions other than the user's position included in the learning data acquired by the data acquisition unit 12 and the unacquired positions included in the learning data acquired by the data acquisition unit 12. Based on the difference from the user position stored, the difference between the detected temperature at the user position contained in the learning data and the set temperature of the indoor unit of the air conditioner, and the set air volume and set air direction of the indoor unit. , the extended data including the difference between the detected temperature at the unacquired position and the set temperature of the indoor unit of the air conditioner, and the set air volume and set air direction of the indoor unit.

10(a) to 10(j) are diagrams showing examples of data obtained from the portable sensor 3 and the control device 2. FIG. FIG. 11 is a diagram showing the positions of the users in FIGS. 10(a) to 10(j).

The position of the user and the detected temperature of the position of the user are obtained by the portable sensor 3 . The set air volume of the indoor unit and the set air direction of the indoor unit are obtained from the control device 2 .

The user's position is represented by polar coordinates (x, y) centering on the indoor unit. The difference between the detected temperature at the user's position and the set temperature of the indoor unit is represented by the temperature difference T at the user's position (x, y). The set air volume of the indoor unit is represented by the air volume W set by the control device 2 at the user's position (x, y). The set wind direction of the indoor unit is represented by the wind direction D set by the control device 2 at the user's position (x, y).

In FIGS. 10A to 10E, the angle of the user's position is a constant value ya, and the distance of the user's position changes as xa, xb, xc, xd, and xe. In FIGS. 10(f) to (j), the angle of the user's position is a constant value yb, and the distance of the user's position changes to xa, xb, xc, xd, and xe.

As shown in FIG. 11, the data on the user's position (xf, ya) is generated from the data on the user's position (xb, ya) and the data on the user's position (xc, ya).

The data extension unit 62 generates data on the user's position (xg, ya) from the data on the user's position (xc, ya) and the data on the user's position (xd, ya). The data extension unit 62 generates data on the user's position (xd, yc) from the data on the user's position (xd, ya) and the data on the user's position (xd, yb).

FIGS. 12A to 12C are diagrams for explaining a method of increasing data.
As shown in FIG. 12(a), the data expansion unit 62 calculates the difference T(xb, ya) between the detected temperature at the user's position (xb, ya) and the set temperature of the indoor unit, the user's position (xc, By linearly interpolating the difference T (xc, ya) between the detected temperature of ya) and the set temperature of the indoor unit, the difference T ( xf, ya).

As shown in FIG. 12(b), the data expansion unit 62 determines the air volume W (xb, ya) set by the control device 2 at the user's position (xb, ya) and By linearly interpolating the air volume W (xc, ya) set by the control device 2, the air volume W (xf, ya) set by the control device 2 is generated at the user's position (xf, ya).

As shown in FIG. 12(c), the data extension unit 62 determines the wind direction D (xb, ya) set by the control device 2 at the user's position (xb, ya) and the wind direction D (xb, ya) at the user's position (xc, ya) By linearly interpolating the wind direction D(xc, ya) set by the controller 2, the wind direction D(xf, ya) set by the controller 2 at the user's position (xf, ya) is generated.

FIGS. 13A to 13C are diagrams for explaining a method of increasing data.
As shown in FIG. 13A, the data extension unit 62 calculates the difference T(xd, ya) between the detected temperature at the user's position (xd, ya) and the set temperature of the indoor unit, and the user's position (xd, ya). By linearly interpolating the difference T (xd, yf) between the detected temperature of the user position (xd, yf) and the set temperature of the indoor unit, the difference T ( xf, yf).

As shown in FIG. 13(b), the data expansion unit 62 determines the air volume W (xd, ya) set by the control device 2 at the user's position (xd, ya), and at the user's position (xd, yb) By linearly interpolating the air volume W (xd, yb) set by the control device 2, the air volume W (xd, yf) set by the control device 2 is generated at the user's position (xd, yf).

As shown in FIG. 13(c), the data extension unit 62 determines the wind direction D (xd, ya) set by the control device 2 at the user's position (xd, ya) and By linearly interpolating the wind direction D(xd, yb) set by the controller 2, the wind direction D(xd, yf) set by the controller 2 at the user's position (xd, yf) is generated.

Modification.
The present disclosure is not limited to the embodiments described above.

(1) FIG. 14 is a diagram showing the hardware configuration of learning device 10, inference device 30, or control device 2. As shown in FIG.

The learning device 10, the reasoning device 30, and the control device 2 can be configured with digital circuit hardware or software for corresponding operations. When the functions of the learning device 10, the reasoning device 30, and the control device 2 are realized using software, the learning device 10, the reasoning device 30, and the control device 2 are connected to the bus 53 as shown in FIG. a processor 51 and a memory 52 connected by a , such that the processor 51 executes a program stored in the memory 52 ;

(2) When there are a plurality of indoor units in the room, they may be interlocked to search for the optimum wind direction and wind speed setting.

It should be considered that the embodiments disclosed this time are illustrative in all respects and not restrictive. The scope of the present disclosure is indicated by the scope of the claims rather than the above description, and is intended to include all changes within the meaning and scope of equivalents of the scope of the claims.

1 air conditioner, 2 control device, 3 portable sensor, 10 learning device, 12, 31 data acquisition unit, 13 model generation unit, 14 reward calculation unit, 15 function update unit, 20 learned model storage unit, 30 reasoning device , 32 reasoning unit, 51 processor, 52 memory, 53 bus.

Claims (5)

A state including a position of a user of an air conditioner, a difference between a detected temperature at the user's position and a set temperature of an indoor unit of the air conditioner, and a set air volume and a set wind direction of the indoor unit in the state. a data acquisition unit that acquires learning data;
Using the learning data, the user's position of the air conditioner and the difference between the detected temperature at the user's position and the set temperature of the indoor unit of the air conditioner are used to determine the set air volume of the indoor unit and the temperature of the indoor unit. a model generation unit that generates a trained model for inferring the set wind direction,
The model generation unit generates the learned model by Q-learning,
The model generating unit increases the reward when the amount of temperature change per unit time at the user's position increases, and increases the reward when the temperature change per unit time at the user's position decreases. A learning device for the control of reducing air conditioners. 2. The learning device for controlling an air conditioner according to claim 1, wherein said data acquisition unit acquires said state from data output from a portable sensor. With respect to unacquired positions other than the user's position included in the learning data acquired by the data acquisition unit, the unacquired position and the user's position included in the learning data acquired by the data acquisition unit. Based on the difference, using the difference between the detected temperature at the user's position and the set temperature of the indoor unit of the air conditioner included in the learning data, and the set air volume and set wind direction of the indoor unit, a data extension unit that generates extended data including the difference between the detected temperature at the unacquired position and the set temperature of the indoor unit of the air conditioner, and the set air volume and set wind direction of the indoor unit;
3. The learning device for controlling an air conditioner according to claim 1, wherein said model generation unit further uses extended data generated by said data extension unit as said learning data. a data acquisition unit that acquires the position of the user of the air conditioner and the state including the difference between the temperature at the user's position and the set temperature of the indoor unit of the air conditioner;
Acquiring the learned model generated by the learning device for controlling the air conditioner according to any one of claims 1 to 3, and using the learned model, the data acquisition unit acquires the an inference unit that infers a set air volume of the indoor unit and a set wind direction of the indoor unit from the state;
A reasoning device for controlling an air conditioner, comprising: 5. The reasoning device for controlling an air conditioner according to claim 4 , wherein said data acquisition unit acquires said state from data output from a portable sensor.

Images