Control method for air conditioner, and device for air conditioner and storage medium

Patent metadata

Field	Value
Publication number	US-11965666-B2
Application number	US-201917600506-A
Country	US
Kind code	B2
Filing date	Dec 16, 2019
Priority date	Apr 1, 2019
Publication date	Apr 23, 2024
Grant date	Apr 23, 2024

Abstract

The disclosure provides a control method and a device for an air conditioner. The method includes: a first reward matrix is constructed according to multiple sets of target operating parameters of an air conditioner, a maximum expected benefit of performing a current action in a current state is calculated based on the first reward matrix and a Q-learning algorithm, wherein the current state is represented by a current indoor environment temperature and a current outdoor environment temperature; target action parameters under the maximum expected benefit are acquired, and operation of the air conditioner is controlled based on second target action parameters, wherein the second target action parameters at least include a second target operating frequency of the compressor, a second target opening degree of the electronic expansion valve and a second target rotating speed of the external fan.

First claim

What is claimed is:

1. A control method for an air conditioner, comprising:
constructing a first reward matrix according to multiple sets of target operating parameters of an air conditioner, wherein each of the multiple sets of the target operating parameters of the air conditioner at least comprises a target indoor environment temperature, a target outdoor environment temperature, a target setting temperature, a target intermediate temperature of an indoor evaporator, a target intermediate temperature of an outdoor condenser, a first target operating frequency of a compressor, a first target opening degree of an electronic expansion valve and a first target rotating speed of an external fan;
calculating a maximum expected benefit of performing a current action in a current state based on the first reward matrix and a Q-learning algorithm, wherein the current state is represented by a current indoor environment temperature and a current outdoor environment temperature, and the current action is represented by a current operating frequency of the compressor, a current opening degree of the electronic expansion valve and a current rotating speed of the external fan; and
acquiring target action parameters under the maximum expected benefit, and controlling operation of the air conditioner based on second target action parameters, wherein the second target action parameters at least comprise a second target operating frequency of the compressor, a second target opening degree of the electronic expansion valve and a second target rotating speed of the external fan.

2. The control method for the air conditioner according to claim 1 , before constructing the first reward matrix according to the multiple sets of the target operating parameters of the air conditioner, further comprising:
collecting air conditioner operating parameters at multiple moments in an actual working condition, wherein the air conditioner operating parameters collected under each of the multiple moments comprise at least one of an indoor environment temperature, an outdoor environment temperature, a setting temperature, an intermediate temperature of the indoor evaporator, an intermediate temperature of the outdoor condenser, an operating frequency of the compressor, an opening degree of the electronic expansion valve and a rotating speed of the external fan; and
performing default parameter filling on the air conditioner operating parameters collected under each of the multiple moments based on a constraint network model, so as to obtain the multiple sets of the target operating parameters of the air conditioner, wherein input of the constraint network model at least comprises the indoor environment temperature, the outdoor environment temperature, the setting temperature, the intermediate temperature of the indoor evaporator and the intermediate temperature of the outdoor condenser, and output of the constraint network model comprises at least the operating frequency of the compressor, the opening degree of the electronic expansion valve and the rotating speed of the external fan.

3. The control method for the air conditioner according to claim 2 , before performing the default parameter filling on the air conditioner operating parameters collected under each of the multiple moments based on the constraint network model, so as to obtain the multiple sets of operating parameters of the target air conditioner, further comprising:
constructing a second reward matrix based on multiple sets of standard operating parameters of the air conditioner, wherein the multiple sets of standard operating parameters of the air conditioner are air conditioner operating parameters satisfying a preset condition;
acquiring training set data and validation set data from the second reward matrix;
training a radial basis function neural network model based on the training set data, wherein the radial basis function neural network model comprises an input layer, a hidden layer and an output layer, and the number of neuron in the hidden layer of the radial basis function neural network model is regulated by an objective loss function;
verifying an accuracy rate of the radial basis function neural network model based on the validation set data; and
using the radial basis function neural network model as the constraint network model in a case that the accuracy rate of the radial basis function neural network model is greater than a preset value.

4. The control method for the air conditioner according to claim 3 , before constructing the second reward matrix based on the multiple sets of the standard operating parameters of the air conditioner, further comprising:
collecting multiple sets of experimental operating parameters of the air conditioner, wherein each of the multiple sets of the experimental operating parameters of the air conditioner at least comprises an experimental indoor environment temperature, an experimental outdoor environment temperature, an experimental setting temperature, an experimental intermediate temperature of the indoor evaporator, an experimental intermediate temperature of the outdoor condenser, an experimental operating frequency of the compressor, an experimental opening degree of the electronic expansion valve and an experimental rotating speed of the external fan;
calculating a reward value of each of the multiple sets of the experimental operating parameters of the air conditioner according to the following formula:
Reward=f(ΔT indoor , ΔT out door ), wherein Reward represents the reward value, f ( represents a reward function, ΔT indoor represents a difference value between the experimental indoor temperature and the experimental setting temperature, ΔT out door represents the difference value between the experimental outdoor temperature and the outdoor temperature under a target working condition; and
taking the set of experimental operating parameters of the air conditioner whose reward value is in a preset scope of the reward value as one set of standard operating parameters of the air conditioner.

5. The control method for the air conditioner according to claim 2 , after controlling the operation of the air conditioner based on the target action parameters, further comprising:
modifying parameters of the first reward matrix based on feedback data, wherein the feedback data is configured to feed back the operation of the air conditioner; and
updating the constraint network model based on the modified first reward matrix whose parameters are modified.

6. The control method for the air conditioner according to claim 1 , wherein calculating the maximum expected benefit of performing the current action in the current state based on the first reward matrix and the Q-learning algorithm comprises:
calculating the maximum expected benefit according to the following formula:
Q ( s,a )= R ( s,a )+Θ*max{ Q ( s next ,a next )}

wherein s represents the current state, a represents the current action, s next represents a next state, a next represents a next action, R(s, a) represents the first reward matrix, Q(s next , a next ) represents a maximum expected benefit obtained through performing the action a next under the state s next , Q(s, a) represents the maximum expected benefit obtained through performing the action a under the state s, and Θ represents a greedy factor.

7. The control method for the air conditioner according to claim 6 , wherein obtaining the target action parameter under the maximum expected benefit comprises:
calculating the target action parameter according to the following formula:
Q ( s,a )=max{ Q ( s,a next )}

wherein s represents the current state, a next represents the next action, Q(S, a next ) represents the maximum expected benefit obtained

Assignees

• Gree Electric Appliances Inc Zhuhai

Inventors

• Tan Jianming
• Li Shaobin
• Song Dechao
• Yue Dong
• Chen Chong
• Luo Xiaoyu
• Deng Jiabi
• Wang Pengfei
• Xiao Wenxuan

Classifications

• F24F11/63 Primary

  Electronic processing · CPC title

• G05B13/027 Primary

  using neural networks only · CPC title

• G05B19/042

  using digital processors (G05B19/05 takes precedence) · CPC title

• G05B2219/2614

  HVAC, heating, ventillation, climate control · CPC title

• F24F11/64

  using pre-stored data · CPC title