Air conditioner control system and control method

Abstract

The embodiment of the application provides an air conditioner control system and a control method, and relates to the technical field of intelligent electrical appliances. The air conditioner control system predicts the setting condition of a user through a machine learning model. The air conditioner control system comprises: an air conditioner system and an algorithm server; the algorithm server is configured to: acquire historical data of the air conditioner and construct a first data set, perform feature analysis on the first data set to obtain a second data set, perform model training on the second data set through a machine learning algorithm, and generate a final model; the historical data comprises user habit historical data; the algorithm server is further configured to: acquire real-time data of the air conditioner, construct a third data set, input the third data set into the final model, and obtain prediction data; the real-time data comprises user setting real-time data; and the algorithm server is further configured to: control the air conditioner according to the prediction data.

Description

Technical Field

[0001] This invention relates to the field of intelligent electrical appliance technology, and in particular to an air conditioning control system and control method. Background Technology

[0002] With the continuous development of household appliance technology, more and more household appliances are entering people's daily lives and work. At the same time, due to the improvement of technology in real life, people have higher demands for the intelligence of household appliances. Therefore, household appliances should also be optimized and upgraded accordingly.

[0003] Taking air conditioners as an example, currently air conditioners can only be turned on and off at different times by setting a timer. However, since the ambient temperature varies at different times of the day, users have different needs for the air conditioner's operating mode and target temperature. They have to set the operating mode and target temperature at different times of the day. The setting parameters for a certain day and the corresponding time periods of several days before and after are the same. This makes it very inconvenient and unintelligent to manually repeat the settings every day. Therefore, how to enable air conditioners to intelligently learn users' air conditioner setting habits, improve the intelligence level of air conditioner control, and meet users' needs has become an existing technical challenge. Summary of the Invention

[0004] This application provides an air conditioning control system and control method, which predicts user settings through a machine learning model.

[0005] In a first aspect, embodiments of this application provide an air conditioning control system, which includes an air conditioning system and an algorithm server; the air conditioning system includes an air conditioner.

[0006] The algorithm server is configured to: acquire historical data of the air conditioner and construct a first dataset; perform feature analysis on the first dataset to obtain a second dataset; and train a model on the second dataset using a machine learning algorithm to generate a final model; the historical data includes historical user habit data.

[0007] The algorithm server is also configured to: acquire real-time data of the air conditioner, construct a third dataset, input the third dataset into the final model, and obtain predicted data; the real-time data includes user-defined real-time data.

[0008] The algorithm server is also configured to control the air conditioner based on the predicted data.

[0009] Based on the above technical solutions, some embodiments of this application provide an air conditioning control system. This control system uses IoT technology and cloud platform big data technology to construct a first dataset from historical air conditioning data, performs feature analysis on the first dataset to obtain a second dataset, and uses machine learning algorithms to simulate and train the second dataset to generate a final model. A third dataset is constructed from real-time air conditioning data collected using IoT technology and cloud platform big data technology, and this third dataset is input into the final model to obtain predicted data. The cloud platform receives the predicted data and controls the air conditioning based on it. By predicting user settings through a machine learning model, the intelligent user experience of smart central air conditioning is greatly enhanced.

[0010] In some embodiments, the algorithm server is further configured to: acquire initial historical data of the air conditioner; the initial historical data of the air conditioner includes: initial historical data of the outdoor environment, initial historical data of the indoor unit operation, initial historical data of user habits, and initial historical data of user settings.

[0011] The initial historical data is merged and filtered to obtain the indoor unit operation history data, the user habit history data, and the user setting history data.

[0012] The initial historical data is merged and filtered to obtain the outdoor environmental historical data.

[0013] The obtained outdoor environment historical data, indoor unit operation historical data, user habit historical data, and user setting historical data are used to construct the first dataset.

[0014] The first dataset is subjected to feature analysis, and data whose feature distribution meets the set requirements are selected to obtain the second dataset.

[0015] In some embodiments, the algorithm server is further configured to: train a model on the second dataset using multiple machine learning methods to generate multiple prediction models; and select a prediction model that meets set conditions as the final model by comparing the performance of the multiple prediction models.

[0016] In some embodiments, the algorithm server is further configured to: obtain the initial historical data of indoor unit operation and the user-set initial historical data according to the device code and indoor unit number of the air conditioner; the algorithm server is further configured to: filter the initial historical data of indoor unit operation; and take the data in the initial historical data of indoor unit operation that meets a first set condition as the historical data of indoor unit operation; the first set condition is that the stable duration of the user-set initial historical data is greater than a first time.

[0017] In some embodiments, the algorithm server is further configured to: obtain initial historical outdoor environmental data of the same region as the user based on the region code; the algorithm server is further configured to: filter the initial historical outdoor environmental data; and select data in the initial historical outdoor environmental data that meets a second set condition as the historical outdoor environmental data; the second set condition is that the interval between the user setting the initial historical data setting time and the collection time of the initial historical outdoor environmental data is less than or equal to a second time.

[0018] In some embodiments, the various machine learning algorithms include: a decision tree-based distributed gradient algorithm and a random forest algorithm.

[0019] In some embodiments, the algorithm server is further configured to: acquire initial real-time data of the air conditioner; merge and filter the initial real-time data to obtain real-time operating data of the indoor unit and real-time user habit data; construct the obtained real-time operating data of the indoor unit and real-time user habit data into the third dataset; input the third dataset into the final model to obtain the prediction data; the prediction data includes: user-set temperature data, user-set fan speed data, and user-set mode data.

[0020] In some embodiments, the air conditioning control system further includes a cloud platform; the algorithm server is further configured to send the prediction data to the cloud platform; the cloud platform is configured to detect whether the prediction data needs to be sent to the air conditioner; if so, the cloud platform sends the prediction data to the air conditioner.

[0021] Secondly, some embodiments of this application provide a control method for an air conditioning control system.

[0022] The control method includes: acquiring historical data of the air conditioner and constructing a first dataset; processing the first dataset to obtain a second dataset; training a model on the second dataset using a machine learning algorithm to generate a final model; the historical data includes user habit history data; acquiring real-time data of the air conditioner and constructing a third dataset; inputting the third dataset into the final model to obtain predicted data; the real-time data includes user-set real-time data; and controlling the air conditioner according to the predicted data.

[0023] The beneficial effects of the control method for the air conditioning control system are the same as those of the air conditioning control system described above, and will not be repeated here. Attached Figure Description

[0024] The accompanying drawings are provided to further understand the technical solutions of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the technical solutions of the present invention, and do not constitute a limitation on the technical solutions of the present invention.

[0025] Figure 1 A schematic diagram of the composition of an air conditioning control system provided in an embodiment of the present invention;

[0026] Figure 2 A flowchart of a control method for an air conditioning control system provided in an embodiment of the present invention;

[0027] Figure 3 A flowchart of a control method for an air conditioning control system during the model training phase, provided as an embodiment of the present invention;

[0028] Figure 4 A flowchart of another air conditioning control system control method in the model training stage provided by an embodiment of the present invention;

[0029] Figure 5 A flowchart of a control method for an air conditioning control system during the model training stage, provided in an embodiment of the present invention;

[0030] Figure 6 A flowchart illustrating another air conditioning control system control method during the model training phase, provided in an embodiment of the present invention;

[0031] Figure 7 A flowchart illustrating another air conditioning control system control method during the model training phase, provided in an embodiment of the present invention;

[0032] Figure 8 This is a schematic diagram illustrating the current outdoor temperature characteristics analysis provided in an embodiment of the present invention;

[0033] Figure 9 This is a schematic diagram illustrating the current outdoor humidity characteristics analysis provided in an embodiment of the present invention;

[0034] Figure 10 This is a schematic diagram illustrating the current outdoor wind force characteristics analysis provided in an embodiment of the present invention;

[0035] Figure 11 This is a schematic diagram of indoor comfort feature analysis provided by an embodiment of the present invention;

[0036] Figure 12 This is a schematic diagram of outdoor comfort characteristic analysis provided by an embodiment of the present invention;

[0037] Figure 13 This is a schematic diagram illustrating the indoor-outdoor temperature difference characteristic analysis provided in an embodiment of the present invention;

[0038] Figure 14A schematic diagram of a machine learning algorithm model provided in an embodiment of the present invention;

[0039] Figure 15 A schematic diagram of another machine learning algorithm model provided in an embodiment of the present invention;

[0040] Figure 16 A schematic diagram illustrating the accuracy of seasonal forecasting provided in an embodiment of the present invention;

[0041] Figure 17 A flowchart of a control method for an air conditioning control system in the setting prediction stage is provided as an embodiment of the present invention;

[0042] Figure 18 A schematic diagram of a fully connected neural network model provided in an embodiment of the present invention;

[0043] Figure 19 This is a flowchart of a control method for an air conditioning control system during the control phase, as provided in an embodiment of the present invention. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0046] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0047] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "connected" and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. Furthermore, when describing pipelines, the terms "connected" and "linked" as used in this invention have the meaning of enabling conduction. The specific meaning needs to be understood in conjunction with the context.

[0048] In embodiments of the present invention, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" or "for example" in embodiments of the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0049] As described in the background section, with the continuous development of household appliance technology, more and more household appliances are entering people's daily lives and work. At the same time, due to the improvement of the level of technology in real life, people have higher demands for the intelligence of household appliances. Therefore, household appliances should also be optimized and upgraded accordingly.

[0050] Taking air conditioners as an example, currently air conditioners can only be turned on and off at different times by setting a timer. However, since the ambient temperature varies at different times of the day, users have different needs for the air conditioner's operating mode and target temperature. They have to set the operating mode and target temperature at different times every day. The setting parameters for a certain day and the corresponding time periods of several days before and after are the same. Currently, users mainly control the air conditioner by operating a wired controller or a mobile APP. This manual repetition every day is inconvenient and not intelligent. Therefore, how to enable the air conditioner to intelligently learn the user's air conditioner setting habits, improve the intelligence level of air conditioner control, and meet the user's needs has become an existing technical challenge.

[0051] To solve this technical problem, an intelligent air conditioning control method is needed to realize the automated and intelligent regulation process of air conditioning, so as to meet the user's needs for air conditioning operation mode and target temperature at different times of the day.

[0052] Based on this, this application provides an air conditioning control system. For example... Figure 1 As shown, the air conditioning control system 1000 includes: an algorithm server 100 and an air conditioning system 200. The air conditioning system 200 includes air conditioners 50. The air conditioners 50 include: multiple indoor units 20 and multiple outdoor units 30 (only one indoor unit and one outdoor unit are shown in the figure).

[0053] Reference Figure 2 The aforementioned algorithm server 100 is configured as follows:

[0054] S1. Obtain historical data of air conditioners and construct the first dataset. Perform feature analysis on the first dataset to obtain the second dataset. Use machine learning algorithms to train the model on the second dataset to generate the final model.

[0055] Historical data includes user habit history data.

[0056] S2. Obtain real-time data from the air conditioner and construct a third dataset. Input the third dataset into the final model to obtain the predicted data.

[0057] Real-time data includes user-defined real-time data.

[0058] S3. Control the air conditioner based on the predicted data.

[0059] It should be noted that the above air conditioning control applies to home multi-split air conditioning users connected to the network.

[0060] This application provides an air conditioning control system in several embodiments. The system constructs a first dataset from collected historical air conditioning data, performs feature analysis on this first dataset to obtain a second dataset, trains the second dataset using machine learning algorithms, and ultimately generates a final model. A third dataset is constructed from real-time air conditioning data collected using IoT technology and cloud platform big data technology. This third dataset is then input into the final model to obtain predicted data. The cloud platform receives the predicted data and controls the air conditioning system accordingly. By predicting user settings through a machine learning model, the intelligent user experience of smart central air conditioning systems is significantly enhanced.

[0061] like Figure 3 As shown, the algorithm server 100 is also configured as follows:

[0062] S11. Obtain the initial historical data of the air conditioner.

[0063] The initial historical data of the air conditioner includes: initial historical data of the outdoor environment, initial historical data of the indoor unit's operation, initial historical data of user habits, and initial historical data of user settings. This initial historical data is collected based on IoT technology and cloud platform big data technology. The algorithm server can store this initial historical data.

[0064] S12. Merge and filter the initial historical data to obtain indoor unit operation history data, user habit history data, and user setting history data.

[0065] S13. Merge and filter the initial historical data to obtain outdoor environmental historical data.

[0066] S14. Construct the first dataset from the obtained historical data of outdoor environment, historical data of indoor unit operation, historical data of user habits, and historical data of user settings.

[0067] S15. Perform feature analysis on the first dataset, filter out the data whose feature distribution meets the set requirements, and obtain the second dataset.

[0068] In some embodiments of this application, the above-mentioned air conditioning control system achieves higher accuracy in model application by merging and filtering.

[0069] Reference Figure 3 The aforementioned algorithm server 100 is also configured as follows:

[0070] S16. Train the model on the second dataset using various machine learning methods to generate multiple prediction models.

[0071] S17. By comparing the performance of multiple prediction models, select the prediction model that meets the set conditions as the final model.

[0072] In some embodiments of this application, the air conditioning control system analyzes user habits through big data and obtains a predictive model based on user habits that is closer to the user, thereby making the model more accurate.

[0073] The initial historical data of the above air conditioners can be found in Table 1 below.

[0074] Table 1. Parameters used by the user to set the prediction function

[0075]

[0076]

[0077] The initial historical data of outdoor environment, the initial historical data of indoor unit operation, the initial historical data of user habits, and the initial historical data of user settings are equivalent to the outdoor environment parameters, indoor unit operation parameters, user habit parameters, and user setting parameters mentioned above.

[0078] The parameters in Table 1 include direct parameters and constructed parameters, with the constructed parameters being constructed from the direct parameters.

[0079] In some embodiments, the variable outdoor comfort index outDI in Table 1 includes two direct parameters: the current outdoor temperature l1 and the current outdoor humidity l2. The formula for calculating the comfort index DI is: DI = T - (0.55 - 0.55f) * (T - 58), where T refers to converting the current outdoor temperature l1 from Celsius to Fahrenheit; and f refers to the current outdoor humidity l2.

[0080] In some embodiments, the variable rec_temp_range_mean in the user habit parameters in Table 1 includes: set temperature iu31Temp, season number season, user ID userId, and timestamp roundTime; user temperature tendency rec_temp_range_mean refers to the value of the user's set temperature in the past week, the past month, and the current season after weighting according to a certain weighting method, that is, rec_temp_range_mean = Average('iu31Temp')by('time_period','weeks','iuId').

[0081] It should be noted that Average('iu31Temp')by('time_period','weeks','iuId') is the average of the user-set temperature over a week; if 'weeks' is changed to 'months', it is the average of the user-set temperature over a month; if 'weeks' is changed to 'season', it is the average of the user-set temperature over the current season.

[0082] Similarly, Average('iu30Wind')by('time_period','weeks','iuId') is the average wind speed set by the user over a week; if 'weeks' is changed to 'months', it will be the average wind speed set by the user over a month; if 'weeks' is changed to 'season', it will be the average wind speed set by the user over the current season.

[0083] `mostly_select('iu29Mode')by('time_period','weeks','iuId')` selects the user's preferred mode over the past week. For example, if the user's preferred mode over the past week is cooling or dehumidifying, and cooling mode is more prevalent than dehumidifying mode, then the user's preferred mode is cooling.

[0084] The timestamp `roundTime` is a Unix timestamp, defined as the total number of seconds since Greenwich Mean Time (GMT) January 1, 1970, 00:00:00. The timestamp does not change with timezone changes, so it remains consistent regardless of your timezone. This avoids errors caused by repeatedly modifying data. To achieve cross-platform compatibility, timestamps can also be used to record time in application systems.

[0085] Among them, the outdoor comfort variable outDI in the outdoor environment parameters is constructed by the outdoor environment parameters themselves, while the user temperature tendency rec_temp_range_mean variable in the user habit parameters uses the season number season in the outdoor environment parameters and the set temperature iu31Temp in the user setting parameters.

[0086] For step S11 above, obtaining the initial historical data of the air conditioner; this initial historical data of the air conditioner is large in volume and diverse in type, but contains little useful information. It is also necessary to mine user usage habits to predict user-set parameters. The libraries used in data processing can be found in Table 2 below:

[0087] Table 2. Libraries used for data processing and their descriptions.

[0088]

[0089] The acquisition and filtering of the initial historical data for the aforementioned air conditioners can be found in [reference needed]. Figure 4 , Figure 5 and Figure 6 .

[0090] Algorithm server 100 is also configured as (see reference) Figure 4 and Figure 5 ):

[0091] S111. Based on the air conditioner's equipment code and indoor unit number, obtain the initial historical data of the indoor unit's operation and the initial historical data of the user settings.

[0092] S121. Filter the initial historical data of the indoor unit.

[0093] S122. Data that meets the first set condition from the initial historical data of indoor unit operation is used as the historical data of indoor unit operation.

[0094] The first setting condition is the user-defined initial historical data stability duration T. W1 Greater than the first time T1, i.e., T W1 >T1.

[0095] In some embodiments, the first time T1 is 10 minutes. For example, if the set temperature iu31Temp in the user setting data obtained with reference to Table 1 is 25°C, and the set temperature iu31Temp changes to another temperature within 10 minutes, then the set temperature iu31Temp is invalid data and will be filtered out; if the set temperature iu31Temp is still 25°C after 10 minutes, then the set temperature iu31Temp is valid data and will be retained.

[0096] Algorithm server 100 is also configured as (see reference) Figure 4 and Figure 6 ):

[0097] S112. Based on the region code, obtain the initial historical outdoor environment data of the same region as the user.

[0098] S131. Filter the initial historical data of the outdoor environment.

[0099] S132. Data that meets the second set condition from the initial historical data of the outdoor environment shall be used as historical data of the outdoor environment.

[0100] The second setting condition is the interval T between the user-defined initial historical data setting time and the initial outdoor environmental historical data collection time. W2 Less than or equal to the second time T2, i.e., T W2 ≤T2.

[0101] In some embodiments, the second time period is 75 minutes. For example, if a user adjusts the air conditioner at 19:00, and the outdoor environmental data of the same area acquired by the air conditioner control system before 17:45 is invalid data, the data will be filtered out; the outdoor environmental data of the same area acquired by the air conditioner control system after 17:45 is valid data, that is, the data within the time period from 17:45 to 19:00 will be retained.

[0102] The air conditioning control system provided in some embodiments of this application merges and filters the initial historical data of the air conditioner, eliminating invalid operation data of some users and unstable data generated by blind operation of users, thereby ensuring the validity of the obtained indoor unit operation history data, user habit history data and outdoor environment history data.

[0103] Of the real-time indoor unit operation data, real-time user setting data, and real-time outdoor environment data obtained above, some data exhibit poor feature distribution and low discriminative power, making it difficult to analyze their effectiveness and hindering algorithm learning. For example... Figure 7 As shown in step S15, perform feature analysis on the first dataset, filter out data whose feature distribution meets the set requirements, and obtain the second dataset; this step also includes the following steps:

[0104] S151. Visualize and analyze the feature distribution in the first dataset, and delete features with poor distribution.

[0105] For a visual analysis of the feature distribution of the second dataset, please refer to [reference needed]. Figures 8-13 .in, Figure 8 It is a visualization analysis chart of the current outdoor temperature l1; Figure 9 It is a visualization analysis chart of the current outdoor humidity L2; Figure 10 It is a visualization analysis chart of the current outdoor wind force L3; Figure 11 This is a visualization analysis chart of indoor comfort level (inDI). Figure 12 This is a visualization analysis chart of outdoor comfort index (outDI). Figure 13 This is a visualization analysis chart of the indoor-outdoor temperature difference temp_range.

[0106] S152. Construct the second dataset from the remaining features of the first dataset.

[0107] The remaining features of the first dataset can be found in Table 3 below.

[0108] Table 3 defines the features constructed during the prediction phase.

[0109]

[0110]

[0111] By constructing features that are more distinctive and better reflect users' individual preferences, we can avoid directly using user ID information, improve the learning effect of machine learning algorithms, and solve the cold start problem for new users.

[0112] In some embodiments, various machine learning algorithms include: a decision tree-based distributed gradient algorithm and a random forest algorithm.

[0113] like Figure 14 As shown, the Light Gradient Boosting Machine (GFCM) based on decision trees uses multiple weak classifiers (decision trees) for iterative training to obtain the optimal model. This model, which integrates multiple weak classifiers, has the advantages of good training effect and is not prone to overfitting. Its characteristics include fast and efficient operation, low memory consumption, high accuracy, and support for large-scale data parallel processing, making it suitable for large-scale data.

[0114] This model uses a depth-constrained leaf-wise growth algorithm, which means that each time, the leaf with the largest splitting gain is found from all the current leaves and split (e.g., ...). Figure 14 (White nodes in the algorithm) until a set depth limit is reached. Compared to the GBDT algorithm, this algorithm can reduce the error generated when selecting features by traversing nodes layer by layer, thus achieving better accuracy. Furthermore, because LightGBM has a maximum depth limit, it ensures high efficiency while preventing the growth of excessively deep decision trees and the resulting overfitting.

[0115] like Figure 15As shown, Random Forest is an algorithm that integrates multiple trees using the Bagging concept of ensemble learning. By building multiple decision trees and fusing them, it obtains a more accurate and stable model, combining the Bagging idea with random feature selection. Random Forest constructs multiple decision trees. When a prediction is needed for a sample, the prediction results of each tree in the forest for that sample are statistically analyzed, and then a voting method is used to select the final result from these prediction results.

[0116] Randomness is reflected in two aspects: one is the random selection of features, and the other is the random selection of samples, so that each tree in the forest has both similarities and differences.

[0117] The growth pattern of each tree in the random forest is as follows:

[0118] (1) N represents the number of training cases (samples), and M represents the number of features.

[0119] (2) Input the number of features m, which is used to determine the decision result of a node on the decision tree; where m should be much smaller than M.

[0120] (3) Sample N times with replacement from N training cases (samples) to form a training set (i.e., bootstrap sampling), and use the unsampled cases (samples) to make predictions and evaluate their errors.

[0121] (4) For each node, randomly select m features. The decision of each node in the decision tree is determined based on these features. Calculate the optimal splitting method based on these m features.

[0122] (5) Each tree will grow fully without being pruned, which may be adopted after a normal tree classifier is built.

[0123] It should be noted that the larger the value of m, the higher the correlation in (3) above, and the stronger the splitting ability in (4). Therefore, m is an important parameter in random forest.

[0124] Reference Figure 15 Following the steps described above, the first decision tree, the second decision tree, and so on up to the Nth decision tree are constructed. All decision trees constitute a random forest, and each decision tree corresponds to a different category. Finally, the random forest integrates all the classification voting results and designates the category with the most votes as the final result.

[0125] Random forests have strong learning capabilities, are not prone to overfitting, have excellent performance, can handle high-dimensional data, and do not require feature selection.

[0126] like Figure 17As shown, the algorithm server 100 is also configured as follows:

[0127] S21. Obtain the initial real-time data of the air conditioner.

[0128] S22. Merge and filter the initial real-time data to obtain real-time indoor unit operation data and real-time user habit data.

[0129] S23. Construct a third dataset from the obtained real-time indoor unit operation data and real-time user habit data.

[0130] S24. Input the third dataset into the final model to obtain the predicted data.

[0131] In some embodiments, the rules for merging and filtering the initial real-time data are the same as those in step S12 above, and will not be repeated here.

[0132] The predicted data includes: user-set temperature data, user-set wind speed data, and user-set mode data.

[0133] For step S24, the third dataset is input into the final model to obtain the predicted data. The final model is... Figure 14 The LightGBM model and Figure 15 The random forest model was used. The model's prediction accuracy was validated using 1000 sets of data from the Central China region exported from the database over one year. The overall prediction accuracy of the model is shown in Table 4.

[0134] Table 4 Overall Accuracy of Model Validation

[0135] Model Temperature prediction accuracy Wind speed prediction accuracy Model prediction accuracy Random Forest 95.26％ 96.51％ 99.82％ LightGBM 93.27％ 94.76％ 96.89％

[0136] Meanwhile, considering the varying usage and data size of air conditioners across seasons, the prediction accuracy for summer might be significantly higher than that for spring, autumn, and winter, leading to an artificially inflated overall prediction accuracy. Therefore, we will continue to calculate the accuracy of user-set predictions for each season, with the prediction accuracy for all four seasons referenced. Figure 16 .

[0137] In conclusion, the Random Forest model has higher prediction accuracy compared to the LightGBM model.

[0138] In some embodiments, the prediction model described above may be a fully connected neural network model.

[0139] Reference Figure 18A fully connected neural network is a network with multiple layers of neurons, with a non-linear activation function between each layer. Historical outdoor environmental data, indoor unit operating data, user habit data, and user setting data are sequentially input into the neural network's input layer; where X1, X2, up to X... n The numbers i1, i2, and up to i represent the input data. n Represents the input layer.

[0140] Then, the predicted values ​​of the target parameters are obtained in the output layer by calculating the activation functions in multiple hidden layers. The loss function is used to calculate the loss between the predicted and true values, and this loss is backpropagated to the neural network through the optimizer, allowing it to correct the weights W between adjacent neurons. Finally, through multiple iterations, the accuracy is improved to the ideal level. The results are then saved to the algorithm server for later use.

[0141] It should be noted that neurons are... Figure 18 The circles in the input layer are i1, i2, and so on up to i n The circle in the hidden layer is H1. (1) H2 (1) H3 (1) Until H m (1) and H1 (s) H2 (s) Until H m (s) The hidden layers include multiple hidden layers, such as H1. (1) H2 (1) H3 (1) Until H m (1) It belongs to the first hidden layer, H1 (s) H2 (s) Until H m (s) It belongs to the S-th hidden layer; the circles in the output layer are O1 to O. t .

[0142] There is a weight W between two adjacent neurons. The preceding neuron is used to generate the following neuron through different weights W, ultimately producing predicted data. The accuracy of the model is judged by how close the predicted data value is to the expected value. The closer the predicted data value is to the expected value, the higher the accuracy of the model; conversely, the closer the predicted data value is to the expected value, the lower the accuracy of the model.

[0143] When the weight W is modified, the subsequent neuron will change, and so on. Through multiple modifications and iterations, the value of the predicted data becomes closer and closer to the expected value, thereby improving the accuracy of the model.

[0144] like Figure 19 As shown, the air conditioning control system 1000 also includes: a cloud platform 300 (see reference). Figure 1 ).

[0145] S31. Transmit the prediction data to the cloud platform.

[0146] S32. The cloud platform detects whether the predicted data needs to be sent to the air conditioner; if yes, proceed to the next step S33; otherwise, go to step S34.

[0147] S33, the cloud platform sends the forecast data to the air conditioner.

[0148] S34, The cloud platform does not take any action.

[0149] S35. The air conditioner receives forecast data and adjusts automatically.

[0150] Some embodiments of this application provide a control method for an air conditioning control system, the control method of which is as follows:

[0151] S1. Obtain historical data of the air conditioner and construct a first dataset; process the first dataset to obtain a second dataset; train a model on the second dataset using a machine learning algorithm to generate a final model; the historical data includes user habit history data;

[0152] S2. Obtain the real-time data of the air conditioner and construct a third dataset. Input the third dataset into the final model to obtain the predicted data. The real-time data includes user-defined real-time data.

[0153] S3. Control the air conditioner according to the predicted data.

[0154] The control method of this air conditioning control system can predict the user's settings in real time through machine learning algorithms, thereby greatly enhancing the intelligent user experience of smart central air conditioning.

[0155] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An air conditioning control system, characterized in that, include: Air conditioning system and algorithm server; the air conditioning system includes air conditioners; The algorithm server is configured to: acquire historical data of the air conditioner and construct a first dataset; perform feature analysis on the first dataset to obtain a second dataset; and train a model on the second dataset using various machine learning algorithms to generate a final model. The algorithm server is also configured to: acquire real-time data of the air conditioner, construct a third dataset, input the third dataset into the final model, and obtain prediction data; The algorithm server is also configured to control the air conditioner based on the predicted data; The algorithm server is also configured to: Obtain the initial historical data of the air conditioner; the initial historical data of the air conditioner includes: initial historical data of outdoor environment, initial historical data of indoor unit operation, initial historical data of user habits, and initial historical data of user settings; The initial historical data is merged and filtered to obtain indoor unit operation history data, user habit history data, and user setting history data; The initial historical data is merged and filtered to obtain outdoor environmental historical data; The obtained outdoor environment historical data, indoor unit operation historical data, user habit historical data, and user setting historical data are used to construct the first dataset. Perform feature analysis on the first dataset, filter out the data whose feature distribution meets the set requirements, and obtain the second dataset; The algorithm server is also configured to: The second dataset is trained using various machine learning algorithms to generate multiple prediction models. By comparing the performance of the various prediction models, the prediction model that meets the set conditions is selected as the final model. The algorithm server is also configured to: Acquire the initial real-time data of the air conditioner; the initial real-time data includes: real-time operating data of the indoor unit, real-time user setting data, and real-time outdoor environment data; The initial real-time data is merged and filtered to obtain real-time indoor unit operation data and real-time user habit data; The obtained real-time indoor unit operation data and real-time user habit data are used to construct the third dataset; The third dataset is input into the final model to obtain the predicted data; the predicted data includes: user-set temperature data, user-set wind speed data, and user-set mode data.

2. The air conditioning control system according to claim 1, characterized in that, The algorithm server is also configured to: Based on the air conditioner's device code and indoor unit number, obtain the indoor unit's initial operating history data and the user-set initial history data; The algorithm server is further configured to: filter the initial historical data of the indoor unit; and use the data in the initial historical data of the indoor unit that meets a first set condition as the historical data of the indoor unit; the first set condition is that the user sets the stable duration of the initial historical data to be greater than a first time.

3. The air conditioning control system according to claim 1, characterized in that, The algorithm server is also configured to: Based on the region code, obtain the initial historical outdoor environment data of the user's region; The algorithm server is further configured to: filter the initial historical data of the outdoor environment; and select data that meets a second set condition from the initial historical data of the outdoor environment as the historical data of the outdoor environment; the second set condition is that the interval between the user-set time of the initial historical data and the collection time of the initial historical data of the outdoor environment is less than or equal to a second time.

4. The air conditioning control system according to claim 1, characterized in that, The various machine learning algorithms include: a decision tree-based distributed gradient algorithm and a random forest algorithm.

5. The air conditioning control system according to claim 1, characterized in that, The air conditioning control system also includes: a cloud platform; The algorithm server is also configured to transmit the prediction data to the cloud platform; The cloud platform is configured to detect whether the prediction data needs to be sent to the air conditioner; if so, the cloud platform sends the prediction data to the air conditioner.

6. A control method for an air conditioning control system, characterized in that, The control method includes: Historical data of the air conditioner is acquired and a first dataset is constructed. Feature analysis is performed on the first dataset to obtain a second dataset. The second dataset is then used to train a model and generate a final model. The real-time data of the air conditioner is obtained, and a third dataset is constructed. The third dataset is then input into the final model to obtain the predicted data. The air conditioner is controlled based on the predicted data; Obtain the initial historical data of the air conditioner; the initial historical data of the air conditioner includes: initial historical data of outdoor environment, initial historical data of indoor unit operation, initial historical data of user habits, and initial historical data of user settings; The initial historical data is merged and filtered to obtain indoor unit operation history data, user habit history data, and user setting history data; The initial historical data is merged and filtered to obtain outdoor environmental historical data; The obtained outdoor environment historical data, indoor unit operation historical data, user habit historical data, and user setting historical data are used to construct the first dataset. Perform feature analysis on the first dataset, filter out the data whose feature distribution meets the set requirements, and obtain the second dataset; The second dataset is trained using various machine learning algorithms to generate multiple prediction models. By comparing the performance of the various prediction models, the prediction model that meets the set conditions is selected as the final model. Acquire the initial real-time data of the air conditioner; the initial real-time data includes: real-time operating data of the indoor unit, real-time user setting data, and real-time outdoor environment data; The initial real-time data is merged and filtered to obtain real-time indoor unit operation data and real-time user habit data; The obtained real-time indoor unit operation data and real-time user habit data are used to construct the third dataset; The third dataset is input into the final model to obtain the predicted data; the predicted data includes: user-set temperature data, user-set wind speed data, and user-set mode data.

Images