Load prediction method and system based on central air conditioner
By constructing a conditional variable matrix and a causal relationship diagram, combining the lagging feature matrix, the final feature matrix is generated, which solves the shortcomings of the existing central air conditioner load prediction method in multivariate dynamic correlation mining, and improves prediction accuracy and model adaptability.
Patent Information
- Application Number
- CN202510167397.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-15
- Publication Date
- 2025-06-03
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing central air conditioner load prediction method lacks in-depth exploration of dynamic correlations between multivariables when building feature, resulting in insufficient prediction accuracy and traditional methods cannot fully explore the complex correlations between multivariables, which limits the model's adaptability.
By collecting central air conditioning environment data, building time series data and load sequence data, defining condition variable matrix, setting sparse optimization targets, filtering significant condition variables, calculating partial correlation coefficients, generating causal relationship diagrams, combining lag feature matrix, generating the final feature matrix for load prediction by machine learning models.
The accuracy of load prediction and the adaptability of the model are improved. Through the synergy between sparse optimization and dynamic significance thresholds, multi-level dimensionality reduction and feature selection are achieved, ensuring the high quality of the final features and enhancing the model's adaptability to complex load changes.
Smart Images
Figure CN120087538A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of air - conditioning load, and particularly to a load prediction method and system based on central air - conditioning. Background Technique
[0002] As an important part of modern buildings, under the background of energy conservation and emission reduction, it is particularly important to optimize the operation efficiency of central air - conditioning. With the progress of data acquisition and transmission technologies, it is more convenient to obtain multi - dimensional and multi - modal data (such as temperature and humidity, energy consumption data, equipment status, etc.) in the building environment, providing richer feature information for load prediction;
[0003] However, in the existing central air - conditioning load prediction, there is a lack of in - depth exploration of the dynamic association between multiple variables during feature construction. Existing methods usually simply superimpose environmental data, failing to fully consider the causal relationship and time - series characteristics between variables, resulting in insufficient prediction accuracy. Secondly, most traditional load prediction methods rely on manual feature selection or simple full - variable input methods, which cannot fully explore the complex associations between multiple variables and easily lead to limitations in model performance. Thirdly, existing causal analysis techniques often use fixed structures or simple correlation analysis, making it difficult to accurately reflect the causal relationship between variables, thus limiting the adaptability of the prediction model to complex systems. Summary of the Invention
[0004] In view of the above existing problems, the present invention is proposed.
[0005] Therefore, the present invention provides a load prediction method based on central air - conditioning to solve the problems that existing methods usually simply superimpose environmental data, fail to fully consider the causal relationship and time - series characteristics between variables, resulting in insufficient prediction accuracy. Secondly, most traditional load prediction methods rely on manual feature selection or simple full - variable input methods, which cannot fully explore the complex associations between multiple variables and easily lead to limitations in model performance. Thirdly, existing causal analysis techniques often use fixed structures or simple correlation analysis, making it difficult to accurately reflect the causal relationship between variables, thus limiting the adaptability of the prediction model to complex systems.
[0006] To solve the above - mentioned technical problems, the present invention provides the following technical solutions:
[0007] In the first aspect, the present invention provides a load prediction method based on central air - conditioning, which includes:
[0008] Collect central air - conditioning environmental data for pre - processing, construct time - series data and load - series data, and generate a conditional variable matrix based on the time - series data by defining conditional variables;
[0009] Set a sparse optimization objective based on the conditional variable matrix, select significant conditional variables by minimizing the conditional correlation between variables, calculate the residual matrix of different variables according to the conditional variable matrix for iterative optimization, screen the key conditional variable set, calculate the partial correlation coefficient and generate a causal relationship diagram;
[0010] Generate a lag feature matrix for the time series data of each variable, calculate the correlation coefficient between each lag variable in the lag feature matrix and the target load variable, screen and retain the lag variables corresponding to the corresponding step size according to the correlation coefficient, and combine the identified parent node set in the causal relationship diagram to generate the final feature matrix;
[0011] Perform load forecasting according to the final feature matrix through a machine learning model and visualize the load curve;
[0012] Perform encrypted and secure transmission of the load forecasting data.
[0013] As a preferred embodiment of the load forecasting method based on central air conditioning according to the present invention, wherein: the environmental data of the central air conditioning is collected and preprocessed to construct time series data and load series data, and the environmental data is collected based on the installation environment of the central air conditioning, including temperature data, humidity data, energy consumption data and equipment switch status data;
[0014] Perform preprocessing on the environmental data, delete the missing values and perform linear interpolation for the missing values, and perform normalization processing on the interpolated data. Construct time series data from the normalized environmental data including temperature data, humidity data and equipment switch status data;
[0015] Based on the energy consumption data, construct load series data according to the time series.
[0016] As a preferred embodiment of the load forecasting method based on central air conditioning according to the present invention, wherein: the generation of the conditional variable matrix by defining conditional variables based on the time series data means using the Tigramite tool based on the Python library to read the time series data X and define the conditional variable set, which is expressed as:
[0017] ;
[0018] where and respectively represent the time series data of the i-th and j-th variables, Z represents the conditional variable matrix, which is the set of all time series variables except and ;
[0019] As a preferred solution of the load prediction method based on the central air conditioner of the present invention, wherein: setting a sparse optimization objective based on the conditional variable matrix, selecting significant conditional variables by minimizing the conditional correlation between variables, calculating the residual matrix of different variables according to the conditional variable matrix for iterative optimization, screening the key conditional variable set, calculating the partial correlation coefficient and generating a causal relationship diagram refers to setting a sparse optimization objective by using a sparse optimization method, and selecting significant conditional variables by minimizing the conditional correlation between variables, which is expressed as:
[0020] ;
[0021] Where represents the regression coefficient vector obtained by sparse optimization, represents two variables and 's residuals, represents the regression coefficient, represents the regularization parameter for controlling the sparsity degree. In the formula, represents the sum of squared residuals, represents the L1 norm regularization;
[0022] Calculate the residual matrix of different variables according to the conditional variable matrix Z, and use the coordinate descent method to iteratively optimize each regression coefficient , which is expressed as:
[0023] ;
[0024] Where represents the jth regression coefficient of the regression coefficient vector, ST represents the sparsification formula, represents the time series data of the jth conditional variable of the conditional variable matrix Z at time t, and T represents the total time;
[0025] Based on the jth conditional variable of the conditional variable matrix Z, perform iterative update, calculate the regression coefficient corresponding to the jth conditional variable in different iteration times, and determine the change amount of the continuous iteration of the regression coefficient in the iterative calculation;
[0026] Based on the historical data, set a minimum value as the convergence threshold. If the change amount of the regression coefficient is less than the convergence threshold, the convergence condition is satisfied;
[0027] Screen out the regression coefficient vector obtained by sparse optimization , where represents the number of columns in the Z matrix. The conditional variables corresponding to the non-zero coefficients are significant key variables, and at the same time, screen out the key conditional variable set , where represents the regression coefficient of the jth conditional variable after sparse optimization;
[0028] Based on the set of key conditional variables, calculate the partial correlation coefficient, denoted as:
[0029] ;
[0030] where represents the partial correlation coefficient at times i and j, represents the conditional covariance, and Var represents the conditional variance;
[0031] Determine the significance level based on historical experience , and according to greater than the significance level values to construct a causal relationship matrix;
[0032] Define nodes based on time series data at different times, and based on the values greater than the significance level in the causal relationship matrix as directed edges, and use as the edge weight, and use a graphical visualization tool to output a causal relationship graph.
[0033] As a preferred solution of the load prediction method based on central air conditioning according to the present invention, wherein: generating a lag feature matrix for the time series data of each variable, calculating the correlation coefficient between each lag variable in the lag feature matrix and the target load variable, screening and retaining the lag variables corresponding to the corresponding step lengths according to the correlation coefficient, combining the set of identified parent nodes in the causal relationship graph to generate a final feature matrix, setting the maximum lag length of the time series data to , based on set the value range of the lag step k ;
[0034] Based on the time series data for the time series data of each variable Generate lag sequence data according to the lag step range , where represents the value of variable at time t−k;
[0035] Based on the lag sequence data, form a new lag feature set , based on the lag feature set, construct a lag feature matrix, and the matrix size is ;
[0036] For each lag variable in the lag feature matrix, calculate the correlation coefficient with the target load variable Y, denoted as:
[0037] ;
[0038] where represents the lag variable The correlation coefficient with the load sequence data Y represents the total length of the time series represents the value at time t in the load sequence data Y represents the mean value of the load sequence data Y represents the lag variable of the mean value
[0039] Based on the calculation of the correlation coefficient, store the correlation coefficients corresponding to all lag steps k as a set
[0040] Calculate the mean and standard deviation of the correlation coefficient set, and use the sum of the mean and standard deviation of the correlation coefficient set as the dynamic significance threshold
[0041] For each lag variable , if the calculated correlation coefficient is greater than or equal to the dynamic significance threshold, retain the corresponding step k, and construct the lag variable with the retained corresponding step as the retained lag feature set
[0042] Identify the parent node set from the causal relationship diagram and merge it with the retained lag feature set to form the final feature matrix, denoted as
[0043] ;
[0044] where represents the final feature matrix represents the parent node set represents the retained lag feature set
[0045] As a preferred solution of the load prediction method based on central air conditioning according to the present invention, wherein: the load is predicted by a machine learning model according to the final feature matrix, the load curve is visualized, and a pre-trained GBDT is used to construct a machine learning model for regression prediction. The machine learning model includes an input layer, a core layer and an output layer
[0046] Input the final feature matrix of the newly collected data into the model and predict the future load value
[0047] Based on the data predicted by the future load value, use a visualization tool to dynamically display the load curve
[0048] As a preferred solution of the load prediction method based on central air conditioning according to the present invention, wherein: the load prediction data is encrypted and securely transmitted. The future load prediction data is organized into a structured file, encrypted using the strong encryption algorithm AES, and a signature is generated for the encrypted data using the digital signature algorithm ECDSA. The encrypted data and the signature are combined into a secure transmission package
[0049] Use the TLS protocol to securely transmit the secure transmission packet. The receiving party decrypts it using the key that matches the sending party, stores the decrypted future load prediction data in the database, and marks its transmission time and signature verification status.
[0050] In a second aspect, the present invention provides a system for a load prediction method based on a central air conditioner, including:
[0051] A data processing module that collects relevant data of the central air conditioner for preprocessing and constructs time series data and load series data;
[0052] A feature construction module that, based on the time series data, defines a set of conditional variables and generates a conditional variable matrix, iteratively optimizes the regression coefficients using the residual matrix by minimizing the conditional correlation between variables, and constructs a set of key conditional variables;
[0053] A causal relationship graph generation module that calculates the partial correlation coefficient, generates a causal relationship graph, and determines the causal relationship between variables;
[0054] A lag feature analysis module that generates a lag feature matrix for the time series data of each variable, calculates the correlation coefficient between each lag variable in the lag feature matrix and the target load variable, and filters the lag variables based on the dynamic significance threshold and retains the key step size;
[0055] A final feature generation module that merges the lag feature set and the set of parent nodes identified in the causal relationship graph to generate a final feature matrix;
[0056] A load prediction module that uses a pre-trained machine learning model to perform regression prediction on the final feature matrix;
[0057] A data secure transmission module that organizes the load prediction data and performs encrypted transmission and secure storage.
[0058] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, it implements any step of the load prediction method based on a central air conditioner as described in the first aspect of the present invention.
[0059] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by the processor, it implements any step of the load prediction method based on a central air conditioner as described in the first aspect of the present invention.
[0060] The beneficial effects of the present invention are as follows: By defining the conditional variable matrix Z, complex interactions in multi-dimensional time series data can be effectively processed. Through the sparse optimization method, the conditional correlation between variables is minimized, and the selected conditional variables have higher significance. By screening the key causal relationships in the causal relationship matrix according to the significance level, the influence of noise or weak correlations on the analysis results is avoided, thereby further improving the reliability of the causal relationship diagram. By setting the maximum lag length and generating lag sequence data, the dynamic change laws of time series data can be fully explored. Through the synergistic effect of sparse optimization and dynamic significance thresholds, multi-level dimensionality reduction and feature selection from conditional variables to time series lag variables are achieved, ensuring the high quality of the final features. Through the pre-trained GBDT model, combined with the causal information and dynamic characteristics in the final feature matrix, the model can adapt to complex load changes in different scenarios and improve the reliability of future load forecasting. BRIEF DESCRIPTION OF THE DRAWINGS
[0061] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.
[0062] Figure 1 It is a schematic flow chart of the load forecasting method based on central air conditioning in Embodiment 1.
[0063] Figure 2 It is a schematic structural diagram of the load forecasting system based on central air conditioning in Embodiment 1. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0064] To make the above objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings of the specification.
[0065] Many specific details are set forth in the following description to facilitate a thorough understanding of the present invention. However, the present invention may be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.
[0066] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or alternative embodiments that exclude each other.
[0067] Embodiment 1, refer toFigure 1 and Figure 2 , which is the first embodiment of the present invention. This embodiment provides a load prediction method based on a central air conditioner, including the following steps:
[0068] S1. Collect the environmental data of the central air conditioner for preprocessing, construct time series data and load series data, and generate a conditional variable matrix based on the time series data by defining conditional variables;
[0069] Preferably, collect the environmental data of the central air conditioner for preprocessing, construct time series data and load series data, and collect environmental data based on the installation environment of the central air conditioner, including temperature data, humidity data, energy consumption data, and equipment switch status data;
[0070] Perform preprocessing on the environmental data, delete missing values and perform linear interpolation for the missing values, and normalize the interpolated data. Construct time series data from the normalized environmental data including temperature data, humidity data, and equipment switch status data;
[0071] Based on the energy consumption data, construct load series data according to the time series.
[0072] By deleting missing values and using linear interpolation to fill data gaps, the integrity of the environmental data is ensured, and the model error caused by incomplete data is reduced. The normalization process eliminates the dimensional differences between temperature, humidity, and equipment switch status data, making the data more consistent in time series modeling, which helps the model learn the rules. Construct the normalized environmental data into time series data, providing complete and time-ordered input for subsequent load prediction, improving the model's ability to capture dynamic environmental changes. By constructing load series data based on energy consumption data, directly associate the historical energy consumption characteristics with time series analysis, ensuring that the prediction model can more accurately reflect energy consumption patterns and trends. The combination of environmental data and load series data provides multi-dimensional input features, enabling the prediction model to capture more feature interaction relationships, enhancing the generalization ability and performance of the prediction. By integrating equipment switch status data, it can better reflect the operating conditions and energy consumption characteristics of the central air conditioner, providing a basis for energy-saving optimization.
[0073] Further, generating a conditional variable matrix based on the time series data by defining conditional variables means using the Tigramite tool based on the Python library to read the time series data X and define a set of conditional variables, expressed as:
[0074] ;
[0075] where and respectively represent the time series data of the i-th and j-th variables, and Z represents the conditional variable matrix, which is except and the set of all time series variables other than
[0076] By using the Tigramite tool, time series data can be quickly read and a set of conditional variables can be defined, providing a concise, clear, and efficient framework for causal relationship analysis. The set of conditional variables Z excludes the direct influence of the target variable and only includes other relevant variables, avoiding interference between target variables, thereby improving the accuracy of causal relationship inference. The way of defining the conditional variable matrix Z can effectively handle complex interactions in multi-dimensional time series data and provide support for mining hidden causal relationships in multi-variable systems.
[0077] S2, set a sparse optimization objective based on the conditional variable matrix, select significant conditional variables by minimizing the conditional correlation between variables, calculate the residual matrix of different variables according to the conditional variable matrix for iterative optimization, screen the key set of conditional variables, calculate the partial correlation coefficient and generate a causal relationship graph;
[0078] Preferably, setting a sparse optimization objective based on the conditional variable matrix, selecting significant conditional variables by minimizing the conditional correlation between variables, calculating the residual matrix of different variables according to the conditional variable matrix for iterative optimization, screening the key set of conditional variables, calculating the partial correlation coefficient and generating a causal relationship graph means using a sparse optimization method to set a sparse optimization objective, selecting significant conditional variables by minimizing the conditional correlation between variables, expressed as:
[0079] ;
[0080] where represents the regression coefficient vector obtained by sparse optimization, indicating the influence of the conditional variable on the residual between variables, represents two variables and 's residuals, represents the regression coefficient, the linear influence coefficient of the conditional variable on the residual, represents the regularization parameter that controls the sparsity level, usually determined by cross-validation. In the formula represents the sum of squared residuals, represents L1-norm regularization, the sparse penalty term of the optimization objective, used to constrain the sparsity of the regression coefficient;
[0081] Calculate the residual matrix of different variables according to the conditional variable matrix Z, and use the coordinate descent method CoordinateDescent to iteratively optimize each regression coefficient expressed as:
[0082] ;
[0083] Among them represents the j-th regression coefficient of the regression coefficient vector, ST represents the sparsification formula, , represents the time series data of the j-th conditional variable of the conditional variable matrix Z at time t, T represents the total time;
[0084] Based on the j-th conditional variable of the conditional variable matrix Z, iterative update is performed. Calculate the regression coefficient corresponding to the j-th conditional variable in different iteration times, and determine the change amount of the continuous iteration of the regression coefficient in the iterative calculation;
[0085] Based on the historical data, set the minimum value as the convergence threshold. If the change amount of the regression coefficient is less than the convergence threshold, the convergence condition is satisfied;
[0086] Screen out the regression coefficient vector obtained through sparse optimization , among which represents the number of columns in the Z matrix. The conditional variables corresponding to the non-zero coefficients are significant key variables, and at the same time, the set of key conditional variables is screened out , among which represents the regression coefficient of the j-th conditional variable after sparse optimization;
[0087] Based on the set of key conditional variables, calculate the partial correlation coefficient, expressed as:
[0088] ;
[0089] Among them represents the partial correlation coefficient between time i and j, represents the conditional covariance, and Var represents the conditional variance;
[0090] Determine the significance level based on historical experience , and according to greater than the significance level value to construct a causal relationship matrix;
[0091] Define nodes based on the time series data at different times, and based on the values greater than the significance level in the causal relationship matrix as directed edges, and as the edge weight, and use a graphical visualization tool to output a causal relationship diagram.
[0092] Through the sparse optimization method, the conditional correlation between variables is minimized. The selected conditional variables have higher significance, which helps to reduce the interference of redundant variables and improve the prediction performance of the model. The L1-norm regularization is used to restrict the sparsity of the regression coefficients, so that the optimization result only retains the conditional variables that have a significant impact on the residual between variables, significantly improving the interpretability of the model. Sparse optimization can quickly screen out the key set of conditional variables in high-dimensional data, avoiding the computational complexity of traditional methods in large-scale data processing. The regression coefficients are gradually updated by the coordinate descent method, and a convergence threshold based on historical data is set to ensure that the optimization process can converge quickly and the calculation process has stability. Calculating the partial correlation coefficient based on the selected key set of conditional variables can more accurately reflect the causal relationship between target variables, avoiding misjudgment caused by the interference of redundant variables or insignificant variables. Screening the key causal relationships in the causal relationship matrix according to the significance level can avoid the influence of noise or weak correlation on the analysis result, thereby further improving the reliability of the causal relationship diagram;
[0093] Sparse optimization selects the conditional variables that have a significant impact on the residual through L1-norm regularization. Combined with the precise construction of the conditional variable matrix, it can effectively filter out redundant or meaningless variables. The selected key set of conditional variables is used to calculate the partial correlation coefficient, avoiding the noise and computational redundancy that may be brought by the full variable set. The calculation of the partial correlation coefficient further reveals the independent causal relationship between target variables. Combined with the accuracy of the key conditional variables, it ensures the accuracy and interpretability of causal analysis. Combined with the conditional variable matrix, the construction of the causal relationship matrix is more reasonable and data-driven, providing a reliable basis for the subsequent construction of the causal relationship diagram. The causal relationship matrix defines the node and edge weights through time series, intuitively reflecting the dynamic causal relationship between variables. The overall technical collaboration not only improves the computational efficiency and accuracy, but also makes the causal analysis result more interpretable, providing comprehensive support for the modeling and decision-making of complex systems.
[0094] S3. Generate a lag feature matrix for the time series data of each variable, calculate the correlation coefficient between each lag variable in the lag feature matrix and the target load variable, screen and retain the lag variables corresponding to the step size according to the correlation coefficient, and combine the identified set of parent nodes in the causal relationship diagram to generate the final feature matrix;
[0095] Preferably, generate a lag feature matrix for the time series data of each variable, calculate the correlation coefficient between each lag variable in the lag feature matrix and the target load variable, screen and retain the lag variables corresponding to the step size according to the correlation coefficient, and combine the identified set of parent nodes in the causal relationship diagram to generate the final feature matrix. Set the maximum lag length of the time series data to , based on Set the value range of the lag step k ;
[0096] Time - series data for each variable based on time - series data Generate lag sequence data according to the lag step range , where represents the variable value at time \(t - k\);
[0097] Form a new lag feature set based on the lag sequence data , and construct a lag feature matrix based on the lag feature set, with the matrix size being ;
[0098] For each lag variable in the lag feature matrix, calculate the correlation coefficient with the target load variable \(Y\), expressed as:
[0099] ;
[0100] where represents the correlation coefficient between the lag variable and the load sequence data \(Y\), represents the total length of the time series, represents the value at time \(t\) in the load sequence data \(Y\), represents the mean of the load sequence data \(Y\), represents the mean of the lag variable ;
[0101] Based on the calculation of the correlation coefficient, store the correlation coefficients corresponding to all lag steps \(k\) as a set;
[0102] Calculate the mean and standard deviation of the correlation coefficient set, and use the sum of the mean and standard deviation of the correlation coefficient set as the dynamic significance threshold;
[0103] For each lag variable , if the calculated correlation coefficient is greater than or equal to the dynamic significance threshold, retain the corresponding step \(k\), and retain the lag variable corresponding to the retained step to construct a retained lag feature set;
[0104] Identify the parent node set from the causal relationship diagram and merge it with the retained lag feature set to form the final feature matrix, expressed as:
[0105] ;
[0106] where represents the final feature matrix, represents the parent node set, represents the retained lag feature set.
[0107] By setting the maximum lag length and generating lag sequence data, the dynamic variation law of time series data can be fully explored, providing key time-dependent characteristics for load forecasting. Using the lag feature matrix and correlation coefficient calculation, it is ensured that only the lag variables strongly correlated with the target load variable Y are retained, reducing the interference of invalid lag variables and improving the efficiency and accuracy of feature selection. By calculating the mean and standard deviation of the correlation coefficient set to obtain the dynamic significance threshold, compared with the fixed threshold, it can better adapt to different load forecasting scenarios, ensuring that the selected lag variables have sufficient significance. By screening lag variables with the dynamic significance threshold, the indiscriminate retention of all lag steps is avoided, reducing the dimension of the final feature matrix and helping to reduce the computational complexity of the model. Merging the set of parent nodes in the causal relationship diagram with the set of retained lag features to generate the final feature matrix can simultaneously retain the causal information and time-dependent characteristics between variables, providing a more comprehensive feature representation for load forecasting. The final feature matrix combines causal relationships and time series characteristics, enabling the load forecasting model to more accurately capture the dynamic and causal associations of the system, thereby improving the prediction accuracy and the generalization ability of the model;
[0108] The generation and screening of lag features form a closed-loop process, not only retaining the key features of time series data but also reducing the introduction of invalid features, ensuring the high quality of the final features. The dynamic threshold can be automatically adjusted according to the specific data characteristics, ensuring the adaptability and rationality of the screening results and having stronger generalization ability than the fixed threshold. Merging the set of parent nodes identified from the causal relationship diagram with the set of retained lag features to form the final feature matrix, this combination incorporates both causal information and time series characteristics, enabling the final feature matrix to reflect both the dynamic dependence of the system and the structured causal associations, improving the expressive ability of the features. The merged final feature matrix integrates the results of multiple technical points to form a high-quality and low-redundancy feature set, ensuring the optimality of the input data for the prediction model;
[0109] Through the synergistic effect of sparse optimization and dynamic significance threshold, multi-level dimensionality reduction and feature selection from conditional variables to time series lag variables are achieved, ensuring the high quality of the final features. The causal influence between variables is highlighted by partial correlation coefficients, while lag features emphasize the dynamic dependence of time series. The combination of the two enables the feature matrix to have both structured causal information and time series characteristics. By combining sparse optimization, lag feature screening, and causal relationship diagrams, a multi-level optimization process is constructed from global conditional variables to local lag variables and then to the final feature matrix, ensuring the scientificity and efficiency of feature selection and enabling the final feature matrix to integrate the global structural information of causal relationships and the dynamic characteristics of time series, significantly enhancing the ability to capture the dynamic changes of complex systems.
[0110] S4. Perform load forecasting based on the final feature matrix through a machine learning model, and visualize the load curve;
[0111] Preferably, perform load forecasting based on the final feature matrix through a machine learning model, visualize the load curve, use a pre-trained GBDT to construct a machine learning model for regression forecasting, and the machine learning model includes an input layer, a core layer, and an output layer;
[0112] Input the final feature matrix of the newly collected data into the model and perform future load value forecasting;
[0113] Based on the data predicted for the future load value, use a visualization tool to dynamically display the load curve.
[0114] Through the pre-trained GBDT model, combining the causal information and dynamic characteristics in the final feature matrix, the model can adapt to complex load changes in different scenarios, improve the reliability of future load forecasting. By dynamically displaying the predicted future load value data as a load curve, it can intuitively reflect the change trend of the future load, facilitating decision-makers to quickly understand the system state and make adjustments. By predicting future load changes, the dynamic load curve provides data support for energy optimization scheduling and energy efficiency management of the central air-conditioning system.
[0115] S5. Encrypt and securely transmit the load forecasting data;
[0116] Preferably, encrypt and securely transmit the load forecasting data, organize the future load forecasting data into a structured file, use the strong encryption algorithm AES to encrypt the file, use the digital signature algorithm ECDSA to generate a signature for the encrypted data, and combine the encrypted data and the signature into a secure transmission package;
[0117] Use the TLS protocol to securely transmit the secure transmission package. The receiving party uses a key matching the sending party to decrypt it, stores the decrypted future load forecasting data in the database, and marks its transmission time and signature verification status.
[0118] Through AES encryption, the future load forecasting data is converted into unreadable ciphertext during transmission, preventing sensitive data from being stolen by unauthorized parties. Using the digital signature algorithm ECDSA to sign the encrypted data ensures that the data has not been tampered with during transmission, and any modification can be quickly detected by the receiving party. By using the TLS protocol, it provides encryption and authentication functions at the transport layer, preventing man-in-the-middle attacks and network sniffing, and improving the security of data transmission. By quickly decrypting and storing the future load forecasting data in the secure transmission package, it ensures that the system can immediately apply the received data to prediction analysis or scheduling decisions.
[0119] This embodiment also provides a system for a load forecasting method based on a central air-conditioning system, including,
[0120] A data processing module that collects relevant data of the central air conditioner for preprocessing and constructs time series data and load series data;
[0121] A feature construction module that, based on the time series data, defines a set of conditional variables and generates a conditional variable matrix, iteratively optimizes the regression coefficients using the residual matrix by minimizing the conditional correlation between variables, and constructs a set of key conditional variables;
[0122] A causal relationship graph generation module that calculates the partial correlation coefficient, generates a causal relationship graph, and determines the causal relationship between variables;
[0123] A lag feature analysis module that generates a lag feature matrix for the time series data of each variable, calculates the correlation coefficient between each lag variable in the lag feature matrix and the target load variable, and filters the lag variables based on the dynamic significance threshold and retains the key step size;
[0124] A final feature generation module that merges the lag feature set and the set of parent nodes identified in the causal relationship graph to generate a final feature matrix;
[0125] A load prediction module that uses a pre-trained machine learning model to perform regression prediction on the final feature matrix;
[0126] A data secure transmission module that sorts out the load prediction data and performs encrypted transmission for secure storage.
[0127] This embodiment also provides a computer device applicable to the case of a load prediction method based on a central air conditioner, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the load prediction method based on a central air conditioner as proposed in the above embodiment.
[0128] The computer device may be a terminal. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.
[0129] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the load prediction method based on a central air conditioner as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, abbreviated as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, abbreviated as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, abbreviated as EPROM), programmable read-only memory (Programmable Red-Only Memory, abbreviated as PROM), read-only memory (Read-Only Memory, abbreviated as ROM), magnetic memory, flash memory, magnetic disk or optical disc.
[0130] In summary, the present invention can effectively handle the complex interactions in multi-dimensional time series data by defining the conditional variable matrix Z. Through the sparse optimization method, the conditional correlation between variables is minimized, and the selected conditional variables have higher significance. According to the significance level, the key causal relationships in the causal relationship matrix are screened to avoid the influence of noise or weak correlations on the analysis results, thereby further improving the reliability of the causal relationship graph. By setting the maximum lag length and generating lag sequence data, the dynamic change rules of time series data can be fully mined. Through the synergistic effect of sparse optimization and dynamic significance threshold, multi-level dimensionality reduction and feature selection from conditional variables to time series lag variables are achieved, ensuring the high quality of the final features. Through the pre-trained GBDT model, combined with the causal information and dynamic characteristics in the final feature matrix, the model can adapt to complex load changes in different scenarios and improve the reliability of future load prediction.
[0131] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.
Claims
1. A load forecasting method based on central air conditioning, characterized in that: include: Collect central air-conditioning environmental data for preprocessing, construct time series data and load series data, define conditional variables based on time series data and generate conditional variable matrix; The sparse optimization target is set based on the conditional variable matrix, and the significant conditional variables are selected by minimizing the conditional correlation between variables. The residual matrix of different variables is calculated according to the conditional variable matrix for iterative optimization, the key conditional variable set is screened, the partial correlation coefficient is calculated, and the causal relationship diagram is generated. Generate a lagged feature matrix for the time series data of each variable, calculate the correlation coefficient between each lagged variable in the lagged feature matrix and the target load variable, screen and retain the lagged variables of the corresponding step length according to the correlation coefficient, identify the parent node set in the causal relationship diagram, and generate the final feature matrix; Use machine learning models to predict load based on the final feature matrix and visualize the load curve; Encrypt and securely transmit load forecast data.
2. The load forecasting method based on central air conditioning according to claim 1, characterized in that: The collected central air-conditioning environmental data is preprocessed to construct time series data and load series data, and environmental data is collected based on the installation environment of the central air-conditioning, including temperature data, humidity data, energy consumption data and equipment switch status data; Preprocess the environmental data, delete missing values, perform linear interpolation on the missing values, and normalize the interpolated data. Construct time series data from the normalized environmental data including temperature data, humidity data, and equipment switch status data. Based on the energy consumption data, load series data is constructed according to the time series.
3. The load forecasting method based on central air conditioning according to claim 2, characterized in that: Defining conditional variables based on time series data to generate a conditional variable matrix refers to using the Tigramite tool based on the Python library to read the time series data X and define a set of conditional variables, which is expressed as: ； in and Represent the time series data of the i-th and j-th variables respectively, Z represents the conditional variable matrix, which is and The set of all time series variables except .
4. The load forecasting method based on central air conditioning according to claim 3, characterized in that: The method of setting a sparse optimization target based on a conditional variable matrix, selecting significant conditional variables by minimizing the conditional correlation between variables, calculating the residual matrix of different variables according to the conditional variable matrix for iterative optimization, screening a set of key conditional variables, calculating partial correlation coefficients and generating a causal relationship diagram refers to setting a sparse optimization target using a sparse optimization method, and selecting significant conditional variables by minimizing the conditional correlation between variables, which is expressed as: ； in represents the regression coefficient vector obtained by sparse optimization, Represents two variables and The residual of represents the regression coefficient, represents the regularization parameter that controls the degree of sparsity, where represents the residual sum of squares, represents L1 norm regularization; Calculate the residual matrix of different variables based on the conditional variable matrix Z, and use the coordinate descent method to iteratively optimize each regression coefficient , expressed as: ； in represents the jth regression coefficient of the regression coefficient vector, ST represents the sparsification formula, represents the time series data of the jth conditional variable of the conditional variable matrix Z at time t, and T represents the total time; Iteratively update the j-th conditional variable based on the conditional variable matrix Z, calculate the regression coefficient corresponding to the j-th conditional variable in different iteration times, and determine the change amount of the regression coefficient in the iterative calculation; The minimum value is set as the convergence threshold based on historical data. If the change in the regression coefficient is less than the convergence threshold, the convergence condition is met; Filter out the regression coefficient vector obtained by sparse optimization ,in Represents the number of columns in the Z matrix, where the conditional variables corresponding to the non-zero coefficients are significant key variables, and the set of key conditional variables is screened out at the same time ,in represents the regression coefficient of the jth conditional variable after sparse optimization; Based on the set of key conditional variables, the partial correlation coefficient is calculated and expressed as: ； in represents the partial correlation coefficient between time i and j, represents conditional covariance, Var represents conditional variance; Determine the significance level based on historical experience , and according to Greater than the significance level The value of constructs the causal relationship matrix; Nodes are defined based on time series data at different times and based on the significance level in the causal relationship matrix. of As a directed edge, As edge weights, a causal graph is output using a graph visualization tool.
5. The load forecasting method based on central air conditioning according to claim 4, characterized in that: The method generates a lagged feature matrix for the time series data of each variable, calculates the correlation coefficient between each lagged variable in the lagged feature matrix and the target load variable, selects and retains the lagged variables of the corresponding step length according to the correlation coefficient, identifies the parent node set in the causal relationship diagram, generates the final feature matrix, and sets the maximum lag length of the time series data to ,based on Set the value range of the lag step size k ; Based on the time series data for each variable Generate lagged series data according to the lag step range ,in Representation variables The value at time t−k; Forming a new set of lagged features based on lagged sequence data , construct the lag feature matrix based on the lag feature set, the matrix size is ; For each lagged variable in the lagged characteristic matrix, the correlation coefficient with the target load variable Y is calculated, expressed as: ； in Represents lagged variable Correlation coefficient with load series data Y, represents the total length of the time series, represents the value of time t in the load series data Y, represents the mean value of the load series data Y, Represents lagged variable The mean of Based on the calculation of the correlation coefficient, the correlation coefficients corresponding to all lag steps k are stored as a set; Calculate the mean and standard deviation of the correlation coefficient set, and use the sum of the mean and standard deviation of the correlation coefficient set as the dynamic significance threshold; For each lagged variable If the calculated correlation coefficient is greater than or equal to the dynamic significance threshold, the corresponding step length k is retained, and the lag variable of the corresponding step length is retained Constructed to preserve the set of lagged features; The parent node set is identified from the causal relationship graph and merged with the retained lag feature set to form the final feature matrix, which is expressed as: ； in represents the final feature matrix, Represents the parent node set, Represents the set of retained lag features.
6. The load forecasting method based on central air conditioning according to claim 5, characterized in that: The load is predicted according to the final feature matrix through the machine learning model, the load curve is visualized, and the machine learning model is constructed using the pre-trained GBDT for regression prediction. The machine learning model includes an input layer, a core layer, and an output layer; The final feature matrix of the newly collected data is input into the model and the future load value prediction is performed; Based on the data of future load value forecast, the load curve is dynamically displayed using visualization tools.
7. The load forecasting method based on central air conditioning according to claim 6, characterized in that: The load forecast data is encrypted and securely transmitted, the future load forecast data is organized into a structured file, the file is encrypted using a strong encryption algorithm AES, a digital signature algorithm ECDSA is used to generate a signature for the encrypted data, and the encrypted data and the signature are combined into a secure transmission package; The secure transmission package is securely transmitted using the TLS protocol. The receiver decrypts it using the key that matches the sender, stores the decrypted future load forecast data in the database, and marks its transmission time and signature verification status.
8. A system for a load forecasting method based on a central air conditioner, based on the load forecasting method based on a central air conditioner according to any one of claims 1 to 7, characterized in that: include, The data processing module collects relevant data of the central air conditioner for preprocessing and constructs time series data and load series data; The feature construction module defines a set of conditional variables and generates a conditional variable matrix based on time series data. It minimizes the conditional correlation between variables and uses the residual matrix to iteratively optimize the regression coefficients to construct a set of key conditional variables. The causal relationship diagram generation module calculates the partial correlation coefficient, generates the causal relationship diagram, and determines the causal relationship between variables; The lag characteristic analysis module generates a lag characteristic matrix for the time series data of each variable, calculates the correlation coefficient between each lag variable in the lag characteristic matrix and the target load variable, screens the lag variables based on the dynamic significance threshold and retains the key step length; The final feature generation module merges the set of lagged features with the set of parent nodes identified in the causal graph to generate the final feature matrix; The load forecasting module uses a pre-trained machine learning model to perform regression forecasting on the final feature matrix; The data security transmission module organizes the load forecast data, encrypts it for transmission, and stores it securely.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the load forecasting method based on central air conditioning described in any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the load forecasting method based on central air conditioning described in any one of claims 1 to 7 are implemented.