Machine learning-based active intelligent control method for air conditioner

Abstract

The present application relates to the technical field of air conditioner control, and relates to a machine learning-based active intelligent control method for an air conditioner. The method comprises: during the starting or running of an air conditioner, collecting environmental data of a space where the air conditioner is located; at least inputting the environmental data into a user analysis model to obtain predicted user operation data; and setting air conditioning control parameters on the basis of the predicted user operation data, wherein the user analysis model is obtained by using historical operation data of a user to train a machine learning model. According to the present application, the training data of the model is real air conditioning control data of the user, and the model can well learn operation information of the user in different environments to form high-dimensional parameter information. Under different environmental conditions, the air conditioner can be actively controlled by simulating habits of the user. The problem that existing air conditioning control strategies cannot accurately realize active intelligent control of air conditioners in a timely manner on the basis of habits of users is solved.

Description

An active intelligent control method for air conditioning based on machine learning

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to the Chinese patent application with application number 202311784050.9 filed with the Patent Office of China on December 22, 2023, entitled “A method for active intelligent control of air conditioners based on machine learning”. The entire contents of the above application are incorporated by reference into this application. Technical Field

[0003] The present application relates to an active intelligent control method for an air conditioner based on machine learning, which belongs to the technical field of air conditioner control; in particular, it relates to the intelligent control of an air conditioner in a smart home scenario. Background Art

[0004] An air conditioner is a household appliance used to regulate indoor air temperature and humidity. Generally speaking, the operation of an air conditioner requires setting the desired target temperature and air speed. The air conditioner will reduce the wind speed or stop operating when the target temperature is reached in the space. These two properties also directly affect the user's physical comfort and energy consumption. In the existing technology, users are required to actively set and adjust the air conditioner set temperature and wind speed based on current environmental factors and their own body sensations. Different users will set different target temperatures and wind speeds based on their own feelings under the same environmental conditions. In home application scenarios, each control of the air conditioner's wind speed and temperature also requires the user to issue corresponding instructions. It will not learn the user's habits, nor will it perform corresponding active control operations.

[0005] The published text of the Chinese invention patent application with publication number CN116878117A discloses an air conditioning control method. The scheme constructs a decision tree and performs decision analysis based on conditional parameters including ambient temperature, user mood and weather, so as to judge the operator's operation control, such as voice control, and whether to directly respond to the voice control or confirm with the user whether to perform the corresponding operation control, thereby avoiding various misoperations or incorrect detection of voice commands that may cause the air conditioning to malfunction.

[0006] This solution has improved the accuracy of air-conditioning control operations to a certain extent and avoided misoperation and malfunction. However, its essence is still to require users to issue operation control instructions based on their feelings. It only judges the credibility of the user's instructions based on environmental conditions and factors. It cannot learn the user's air-conditioning usage habits and will not perform active operation control.

[0007] Furthermore, Chinese invention patent application publication number CN105020838A discloses a method for controlling an air conditioner. This method automatically determines the air conditioner's output temperature based on indoor and outdoor ambient temperatures and historical output temperature values ​​determined based on historical user operations, eliminating the need for manual adjustment and simplifying user operations.

[0008] However, this solution only calculates the current air-conditioning output temperature through a preset, fixed-coefficient simple relationship function involving indoor and outdoor temperatures, combined with the user-set temperatures of the past few days. It has a weak learning ability for user habits and has a lag. The user's intervention operation can only be gradually reflected in the output temperature settings of the next few days through the average, which is not timely and accurate enough.

[0009] These existing air conditioning control strategies and solutions have poor feedback mechanisms and are unable to accurately and timely self-correct and modify strategies based on user input. Furthermore, they are unable to learn and adapt to user habits at different times of the day, and adopt different control strategies for different time periods.

[0010] Summary of the Invention

[0011] The purpose of this application includes, for example, providing an active intelligent control method for air conditioning based on machine learning to solve the problem that existing air conditioning control strategies cannot accurately and timely realize active intelligent control of air conditioning according to user habits.

[0012] The present application provides a technical solution for an active intelligent control method for an air conditioner based on machine learning. The method includes: collecting environmental data of the space where the air conditioner is located during the startup or operation of the air conditioner, at least inputting the environmental data into a user analysis model to obtain predicted user operation data, and setting air conditioner control parameters based on the predicted user operation data; the user analysis model is obtained by training a machine learning model using historical user operation data.

[0013] Optionally, the current time is also input into the user analysis model to obtain the user operation data predicted at the current time; the user historical operation data used in the training of the user analysis model has a time label.

[0014] Optionally, the environmental data includes air temperature and / or air humidity data.

[0015] Optionally, the user operation data includes a set temperature of the air conditioner and / or a set wind speed of the air conditioner.

[0016] Optionally, the user analysis model is obtained by training a decision tree model using user historical operation data.

[0017] Optionally, after the air conditioning control parameters are set according to the user operation data, if the user actively controls and modifies the air conditioning control parameters within a set time interval, the machine learning model is retrained according to the modified air conditioning control parameters.

[0018] Optionally, if the environmental data cannot be collected, the most frequently used user operation data at the same time within a set time range in the historical operation data is used as the predicted user operation data.

[0019] Optionally, users are also differentiated, and a correlation analysis is performed between the user operation data and the corresponding environmental data based on the user's historical operation data. For users whose correlation is lower than a set level, the most frequently used user operation data at the same time within a set time range in the historical operation data is used as the predicted user operation data.

[0020] Optionally, Spearman's correlation coefficient was used for correlation analysis.

[0021] Optionally, the current perceived temperature is also calculated based on the air temperature and air humidity in the environmental data. If the conditions met include that the current ambient temperature and the current perceived temperature differ by more than a set degree, the air conditioning wind speed is directly set to the maximum wind speed.

[0022] Optionally, the method includes a user operation data integration step, in which the user operation data is filtered, and data with more than 500 operations per day are filtered out by time.

[0023] Optionally, in the user operation data integration step, for user operation data that changes multiple times within 10 minutes, the user operation data of the latest change is extracted.

[0024] Optionally, in the user operation data integration step, the user operation data is classified by time, and a full day of 24 hours is divided into 9 time segments, each time segment includes 3 hours.

[0025] Optionally, the environmental data includes the concentration of carbon dioxide in the air.

[0026] Optionally, the air conditioning control parameters are updated once a day based on user operation data.

[0027] The beneficial effects of this application include, for example:

[0028] The method of the present application inputs current environmental data, including air temperature and humidity, and the current time of day, into a user analysis model that pre-uses historical data and a machine learning decision tree when the air conditioner is started, or at regular intervals during operation. The model then outputs predicted user operation data, including the air conditioner set temperature and set wind speed, and sets the air conditioner control parameters based on this predicted user operation data. This application learns from users' air conditioner usage habits in different environments, predicts and mimics user preferences, and sets air conditioner control parameters when the air conditioner is started or during operation. The application also has a mature feedback learning mechanism that can accurately predict user operations, enabling intelligent and proactive control of smart home air conditioners and improving user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] FIG1 is a flowchart of a user operation data integration process according to an embodiment of the present application;

[0030] FIG2 is a flowchart of constructing a user analysis model in an embodiment of the present application;

[0031] FIG3 is a flow chart of the air conditioning control logic of the air conditioning active intelligent control method based on machine learning in an embodiment of the present application;

[0032] FIG4 is a logic flow of air conditioning control considering the final perceived temperature in an embodiment of the present application. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solutions and advantages of this application more clear, this application is further described in detail below with reference to the accompanying drawings and embodiments.

[0034] The idea of ​​this application is to first analyze the user's historical air conditioning operation records, analyze the common temperature and common wind speed in each time period, and collect the air information corresponding to the time in the operation record for the room with an air box that collects environmental data (capable of collecting air data such as air temperature and humidity), including temperature, humidity, carbon dioxide concentration, etc., and use the Spearman correlation algorithm to perform correlation analysis between temperature, humidity and set temperature. The data with correlation is trained using a decision tree classification model. The training data of the model is the user's real air conditioning control data. The model can well learn the user's operation information in different environments and form high-dimensional parameter information. It can achieve active control of the air conditioner by imitating the user's habits under different environmental conditions.

[0035] The present invention provides an active intelligent air conditioning control method based on machine learning, including the following contents:

[0036] 1. First, user operation data integration is performed. The user operation data integration process is shown in Figure 1 and includes the following steps.

[0037] (1) Filter the user's operation data and filter out data with more than 500 operations per day by time to prevent excessive redundant data from affecting user behavior analysis.

[0038] (2) For control data that changes multiple times within 10 minutes, the data after the last change is used as the main data. That is, the change data is rolled over at a preset time of 10 minutes, and the temperature and wind speed control data of the last change are extracted. The data are integrated to form a user operation record table.

[0039] (3) In the user operation record table, for control data (including temperature control data and wind speed control data) containing null values, the previous control record is searched and used as the corresponding control data. For example, if an air conditioning operation instruction does not contain wind speed or temperature information, the user operation record table is searched forward until the most recent temperature or wind speed control data is found, which is used as the temperature or wind speed control data for the operation instruction.

[0040] After obtaining the user operation record table, a user analysis model for obtaining the current user operation data is constructed for different users.

[0041] 2. For users without air boxes (or rooms without air boxes), operational data modeling is performed based on operational data over different time periods. The air box can collect environmental data in the room, such as air temperature, air humidity, and carbon dioxide concentration in the air.

[0042] Before modeling, the operation data in the user operation record table in step 1 is classified by time. In this embodiment, a full day of 24 hours is divided into the following 9 time segments, each of which consists of 3 hours. At the same time, for the convenience of program writing, the time segments are also converted to minutes:

[0043] First, all operation data are sorted in descending order by time. In each time period, data are aggregated by time to form a set A. ij , i represents the time period (segment label), i is from 0 to 8; j represents the control data type in the time period, j is a positive integer greater than 0.

[0044] To A ij Content statistical analysis is performed by time period to analyze the recently frequently used air conditioner setting temperature and air conditioner setting wind speed, which are used as the user's commonly used temperature and wind speed in this time period, and used as a user analysis model for users who do not have air boxes.

[0045] 3. For users with air boxes (or rooms where a user has an air box), correlation analysis is performed between environmental data (air temperature and air humidity collected by the air box) and user operation data (air conditioner set temperature set by the user).

[0046] In this embodiment, the Spearman correlation coefficient is used to perform a correlation analysis on the user's ambient temperature, humidity, and the user-set temperature.

[0047] The calculation of the Spearman correlation coefficient and the correlation determination belong to the prior art. In this embodiment, the relevant steps are only briefly introduced as follows:

[0048] (1) For each temperature or humidity, sort it according to the size of the value and assign a level;

[0049] (2) If there are the same temperature or humidity, they are assigned an average level;

[0050] (3) For each pair of temperature or humidity observations, calculate the difference in their ranks in the two variables;

[0051] (4) For each pair of observations, sum the squares of the rank differences;

[0052] (5) Calculate the Spearman correlation coefficient using the following formula:

[0053] where rs is the Spearman correlation coefficient, d is the rank difference, and n is the total number of observations.

[0054] The corresponding relationship between the Spearman correlation coefficient value and the correlation is shown in the following table:

[0055] 4. For users with a certain correlation between their environmental data and user operation data (hereinafter referred to as "correlated users"), a user analysis model based on a machine learning decision tree is trained to predict user operation parameters based on the environmental data. In this embodiment, strong correlation and above are defined as having a certain correlation.

[0056] The following describes this step by taking the time period, air temperature and air humidity as environmental data, and the user-set temperature and wind speed (air conditioner set temperature and wind speed) as user operation data as an example.

[0057] Using a machine learning decision tree algorithm, we predict the user's set temperature based on time period, indoor temperature, and humidity. The decision tree divides nodes based on "maximizing information gain." Information gain calculation is a state-of-the-art technology. The following briefly explains the steps involved:

[0058] Input: training data set D and attribute a (where each attribute a has V possible values ​​{a 1 ,a 2 ,a 3 ,…,a v}), attribute a is the input and output parameter of the decision tree algorithm, and in this embodiment includes 5 (i.e., V=5), namely, the grouped time periods, indoor air temperature and air humidity in the user operation data, and the air-conditioning temperature and air-conditioning wind speed set by the corresponding user.

[0059] Output: Information gain Gain(D,a) of attribute a for training data set D.

[0060] (1) Assume that the proportion of the i-th class samples in the sample set of the training data set D is p i (i=1,2,3,…,|a|), we need to calculate the information entropy first.

[0061] (2) Information entropy formula:

[0062] Where n is the number of operation data of a user with environmental information in the training dataset D.

[0063] (3) Calculate the conditional information entropy of the discrete feature a for the training data set D. The conditional information entropy formula is:

[0064] (4) Calculate the information gain. The information gain formula is: Gain(D,a)=Ent(D)-Ent(D|a)

[0065] For each relevant user, a decision tree model of the user is constructed as the user analysis model to fully learn the relationship between the user's set temperature, wind speed and ambient temperature and humidity. Then, the learned user analysis model is used to predict the set temperature and wind speed that imitates the user's habits.

[0066] For non-correlated users, a user analysis model of the user's common temperature and wind speed in different time periods can be constructed in the same way as for users without air boxes, and this can be used as the user analysis model for non-correlated users.

[0067] The process of establishing a user analysis model for users with environmental data is shown in Figure 2. After obtaining the user operation table records (user operation record table), for users with too few operation records to effectively train the decision tree model, based on a small number of recent operation records, a similar method is adopted as for users without environmental data (no air box), and the air conditioner is set using the common temperature and wind speed in the same time period in the recent operation records. If there are no recent operation records, no processing is done, and the air conditioner temperature and wind speed can be set according to the existing technology, for example, the air conditioner is set according to the set temperature and wind speed when it was last turned off; if there are enough valid operation records, the decision tree model is trained according to the above steps 3 and 4. The environmental data of the training model can also be further increased with carbon dioxide concentration on the basis of air temperature and air humidity to obtain a user analysis model.

[0068] 5. As shown in FIG3 , the air conditioning control logic of the air conditioning active intelligent control method based on machine learning of the present application.

[0069] (1) For users with environmental data (i.e., users with air boxes), obtain the current control time period Time, the current environment's air temperature T, and the air humidity H. Input Time, T, and H into the corresponding user's decision tree model as the user analysis model, and obtain the model output temperature as the predicted user-set temperature.

[0070] (2) For users (or rooms) without environmental data and users without correlation, the temperature and wind speed are set to the common temperature and wind speed in the corresponding time period in the user analysis model (the common temperature and wind speed in the corresponding time period are the output temperature and wind speed of the user analysis model).

[0071] (3) Correct the set temperature. If the user actively inputs the temperature or wind speed, the air conditioner set temperature and wind speed will be modified to the temperature and wind speed input by the user. If the user does not set the air conditioner set temperature and wind speed, the air conditioner set temperature and wind speed will be set to the output temperature and wind speed of the user analysis model.

[0072] As an optional implementation, the active intelligent air conditioning control method of the present application is also updated at set intervals (e.g., daily) based on user habits. If the user is dissatisfied with the control results given by the algorithm and makes corresponding modifications within a preset time interval (e.g., 10 minutes), the model is retrained based on the modified data to generate new results. The emergence of the above feedback mechanism can correct the model to a certain extent and improve the accuracy of the algorithm.

[0073] The machine learning-based active intelligent air conditioning control method of the present application analyzes the user's commonly used temperature and wind speed in the corresponding environment and time period through a user analysis model, calculates the temperature and wind speed that need to be controlled based on the temperature and humidity of the environment according to the algorithm model, directly supplements the temperature and wind speed attributes of the air conditioner that are missing when the user turns on the air conditioner, and performs NLP (natural language processing) active control, reducing user operations and increasing the smart home active control experience.

[0074] Based on the above embodiments, this embodiment further obtains the final perceived temperature based on the temperature and humidity of the environment for users with environmental data.

[0075] (1) When the input is the current ambient temperature in Celsius, first convert Celsius to Fahrenheit.

[0076] The relationship between Fahrenheit temperature F and Celsius C is: F = C × 9 / 5 + 32;

[0077] (2) Rough calculation of the perceived temperature HI0: HI0 = 0.5 × (T + 61 + (T - 68) × 1.2 + RH × 0.094)

[0078] (3) When the rough perceived temperature is greater than 80°F, the perceived temperature is accurately calculated using the following formula: HI1 = -42.379 + (2.04901523 × T) + (10.14333127 × RH) - (0.22475541 × T × RH) - (6.83783e-3 × T 2 )-(5.481717e-2×RH 2 )+(1.22874e-3×T 2 ×RH)+(8.5282e-4×T×RH 2 )-(1.99e-6×T 2 ×RH 2 )

[0079] Among them, HI1 is the accurately calculated body temperature, T is the ambient temperature, and RH is the relative humidity.

[0080] (4) When the humidity is less than 13% and the Fahrenheit temperature is (80, 112), calculate the somatosensory deviation factor TI1 and obtain the final somatosensory temperature HI: HI=HI1-TI1

[0081] (5) When the humidity is greater than 85% and the Fahrenheit temperature is (80, 87), calculate the somatosensory deviation factor TI2 and obtain the final somatosensory temperature HI: HI=HI1+TI2

[0082] As shown in Figure 4, in order to achieve rapid cooling, the air conditioner fan speed is set to the highest speed based on the user's habitual temperature, ambient temperature, and final perceived temperature. The air conditioner fan speed is set to high speed when the following conditions are met:

[0083] (1) When the current month is July or August, and it is during the daytime and not during sleeping hours;

[0084] (2) The user's historical operation records in the same time period contain three or more high wind setting operations;

[0085] (3) The current ambient temperature and the final perceived temperature differ by more than 3 degrees;

[0086] When the above three conditions are not met, the wind speed is set to the output wind speed of the user analysis model. Industrial Applicability

[0087] The present application provides an active intelligent control method for air conditioning based on machine learning, which can realize active control of the air conditioning by imitating the user's habits under different environmental conditions, and can realize active intelligent control of the air conditioning according to the user's habits accurately and timely.

[0088] In addition, it can be understood that the machine learning-based active intelligent air-conditioning control method of the present application is reproducible and can be widely used in the field of air-conditioner control technology.

Claims

1. An active intelligent air conditioning control method based on machine learning, characterized in that: During the startup or operation of the air conditioner, environmental data of the space where the air conditioner is located is collected, and at least the environmental data is input into the user analysis model to obtain predicted user operation data, and the air conditioner control parameters are set according to the predicted user operation data; the user analysis model is obtained by training the machine learning model using the user's historical operation data.

2. The method for active intelligent air conditioning control based on machine learning according to claim 1, characterized in that: The current time is also input into the user analysis model to obtain the predicted user operation data at the current time; the user historical operation data used in the training of the user analysis model has a time label.

3. The method for active intelligent air conditioning control based on machine learning according to claim 1, characterized in that: The environmental data include air temperature and / or air humidity data.

4. The method for active intelligent air conditioning control based on machine learning according to claim 1, characterized in that: The user operation data includes a set temperature of the air conditioner and / or a set wind speed of the air conditioner.

5. The method for active intelligent air conditioning control based on machine learning according to claim 1, characterized in that: The user analysis model is obtained by training a decision tree model using user historical operation data.

6. The method for active intelligent air conditioning control based on machine learning according to claim 1, characterized in that: After the air conditioning control parameters are set according to the user operation data, if the user actively controls and modifies the air conditioning control parameters within the set time interval, the machine learning model is retrained according to the modified air conditioning control parameters.

7. The method for active intelligent air conditioning control based on machine learning according to claim 1, characterized in that: If the environmental data cannot be collected, the most frequently used user operation data at the same time within the set time range in the historical operation data is used as the predicted user operation data.

8. The method for active intelligent air conditioning control based on machine learning according to claim 1, characterized in that: Users are also differentiated, and a correlation analysis is performed between the user operation data and the corresponding environmental data based on the user's historical operation data. For users whose correlation is lower than a set level, the most frequently used user operation data at the same time within a set time range in the historical operation data is used as the predicted user operation data.

9. The method for active intelligent air conditioning control based on machine learning according to claim 8, characterized in that: Correlation analysis was performed using the Spearman correlation coefficient.

10. The method for active intelligent air conditioning control based on machine learning according to claim 1, characterized in that: The current perceived temperature is also calculated based on the air temperature and air humidity in the environmental data. If the conditions met include that the current environmental temperature and the current perceived temperature differ by more than a set degree, the air conditioning wind speed is directly set to the maximum wind speed.

11. The method for active intelligent air conditioning control based on machine learning according to claim 1, characterized in that: The method comprises a user operation data integration step, in which the user operation data is filtered, and data with more than 500 operations per day are filtered out according to time.

12. The method for active intelligent air conditioning control based on machine learning according to claim 11, characterized in that: In the user operation data integration step, for user operation data that changes multiple times within 10 minutes, the user operation data of the latest change is extracted.

13. The method for active intelligent air conditioning control based on machine learning according to claim 11, characterized in that: In the user operation data integration step, the user operation data is classified by time, and a whole day of 24 hours is divided into 9 time segments, each time segment includes 3 hours.

14. The method for active intelligent air conditioning control based on machine learning according to claim 1, characterized in that: The environmental data includes the concentration of carbon dioxide in the air.

15. The method for active intelligent air conditioning control based on machine learning according to any one of claims 1 to 14, characterized in that: The air conditioning control parameters are updated once a day based on user operation data.

Images