Regional air conditioner load prediction method and system based on large time sequence model

Abstract

The invention discloses a regional air-conditioning load prediction method based on a time sequence large model. The method comprises the following steps: acquiring historical air-conditioning load data and key variables related to air-conditioning loads; the air conditioner load data and the key variables are processed through a Prompt technology, and a processed data sequence is obtained; constructing a time sequence prediction large model based on a UniTS model; training and testing the model through the processed data sequence to obtain a trained time sequence prediction large model; and regional air conditioner load prediction is carried out through the trained time sequence prediction large model. According to the method, a large time sequence prediction model based on a Prompt + UniTS model is constructed, efficient processing and intelligent analysis of multivariable data sequences among air conditioner load data and various related data related to the air conditioner load data are achieved, the accuracy and reliability of regional air conditioner short-term prediction are improved, and powerful support is provided for safe and efficient operation of a comprehensive energy system.

Description

Technical Field

[0001] The invention relates to power consumption control, and in particular to a method and system for predicting regional air-conditioning load based on a large time series model. Background Art

[0002] Air conditioning energy consumption accounts for a significant share of building energy consumption. According to authoritative data, especially in summer, in some densely populated metropolises, indoor air conditioning power load even accounts for more than half of the total power load. This shows that the use of air conditioning has an impact on urban power consumption that cannot be ignored. Large-scale air conditioning start and stop bring new peak loads, so when conducting power regulation, short-term prediction and precise regulation of indoor cooling loads are particularly critical. In the field of air conditioning power load prediction, existing technologies all use a single prediction model with low prediction accuracy and poor generalization ability. Summary of the invention

[0003] Purpose of the invention: In view of the above shortcomings, the present invention provides a regional air-conditioning load forecasting method and system based on a large time series model with high accuracy and generalization capability of air-conditioning load forecasting.

[0004] Technical solution: To solve the above problems, the present invention adopts a regional air conditioning load prediction method based on a time series large model, including:

[0005] Obtain historical air conditioning load data and key variables related to air conditioning load;

[0006] The air conditioning load data and key variables are processed by Prompt technology to obtain the processed data sequence;

[0007] Build a time series prediction model based on the UniTS model; train and test the model through the processed data sequence to obtain a trained time series prediction model;

[0008] Regional air conditioning load forecasting is performed using a trained time series forecasting model.

[0009] Furthermore, the step of obtaining key variables related to air conditioning load includes:

[0010] Through Spearman correlation analysis, the key variables related to air conditioning load are screened out, and the Spearman correlation coefficient p calculation formula is:

[0011]

[0012] in, is the rank difference, and n is the number of data.

[0013] Furthermore, the step of obtaining historical air-conditioning load data includes: preprocessing the obtained historical air-conditioning load data, the preprocessing includes smoothing the historical air-conditioning load data using a local weighted scatter plot smoothing algorithm; the local weighted scatter plot smoothing algorithm uses polynomial weighted fitting and performs error estimation through the least squares method; the local weighted scatter plot smoothing algorithm uses a tricubic weight function.

[0014] Furthermore, the large time series prediction model based on the UniTS model includes N stacked UNITS blocks and a CLS component, and the UNITS block includes a sequence multi-head self-attention MHSA, a variable MHSA, a dynamic multi-layer perceptron MLP and a gate module; the large time series prediction model based on the UniTS model converts the original sequence into three types of token tags, including sequence token, prompt token and task token, the sequence token divides the original sequence into multiple patch fragments through non-overlapping windows, and each patch fragment is mapped by a dynamic multi-layer perceptron MLP to obtain a corresponding embedding to characterize the sequence characteristics of the original sequence; the prompt token is used to absorb contextual information in the original sequence according to task characteristics; the task token includes a mask token and a cls token, the mask token is used to uniformly process time series prediction and filling tasks to generate prediction results, and the cls token is used for time series classification tasks.

[0015] Furthermore, the method of training and testing the model through the processed data sequence to obtain a trained time series prediction large model includes: using the Gray Wolf Optimization Algorithm to optimize the network hyperparameters in the time series prediction large model with the goal of minimizing the difference between the air conditioning load prediction value output by the time series prediction large model and the actual value of the air conditioning load.

[0016] Furthermore, the dynamic multi-layer perceptron MLP uses a dynamic linear operator Dylinear to process the original MLP weights, performs bilinear interpolation according to the length of the input vector, and generates MLP weights that match the input dimension. The Dylinear operation is as follows:

[0017] DyLinear(z s ; w) = W interp z s ; W interp =Interp(w)

[0018] Among them, Zs is a given sequence label, w is the weight, Interp is the bilinear interpolation.

[0019] The present invention also adopts a regional air conditioning load prediction system based on a time series large model, comprising:

[0020] A data acquisition module is used to obtain historical air conditioning load data and key variables related to air conditioning load;

[0021] The data processing module is used to process the air conditioning load data and key variables through the prompt technology to obtain the processed data sequence;

[0022] The model building module is used to build a large time series prediction model based on the UniTS model; the model is trained and tested by the processed data sequence to obtain a trained large time series prediction model;

[0023] The prediction module is used to predict regional air conditioning load through the trained time series prediction model.

[0024] Furthermore, the data acquisition module is also used to screen out key variables related to the air conditioning load through Spearman correlation analysis. The calculation formula of the Spearman correlation coefficient p is:

[0025]

[0026] in, is the rank difference, and n is the number of data.

[0027] Furthermore, the data acquisition module is also used to preprocess the acquired historical air-conditioning load data, and the preprocessing includes smoothing the historical air-conditioning load data using a local weighted scatter plot smoothing algorithm; the local weighted scatter plot smoothing algorithm uses polynomial weighted fitting and performs error estimation through the least squares method; the local weighted scatter plot smoothing algorithm uses a tricubic weight function.

[0028] The present invention also adopts a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0029] The present invention also adopts a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are implemented.

[0030] Beneficial effects: Compared with the prior art, the significant advantage of the present invention is that it constructs a large time series prediction model based on the Prompt+UniTS model, realizes efficient processing and intelligent analysis of multivariate data sequences between air-conditioning load data and various related data, improves the accuracy and reliability of regional air-conditioning short-term predictions, and provides strong support for the safe and efficient operation of integrated energy systems. By introducing the Spearman correlation analysis method to screen key variables and the LOWESS non-parametric learning method to reduce data noise, the prediction accuracy and efficiency are effectively improved. The optimization of network hyperparameters by the Grey Wolf Optimization Algorithm (GWO) ensures the stability and accuracy of the model when facing different data sources and sequence lengths. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 Schematic diagram of the prediction method of the present invention.

[0032] Figure 2 This is a correlation analysis of key variables related to air conditioning load in the present invention.

[0033] Figure 3 This is a comparison chart of the initial prediction results of air conditioning load in the present invention.

[0034] Figure 4 This is a comparative analysis diagram of the air conditioning load prediction results after the prediction model is optimized in the present invention. DETAILED DESCRIPTION

[0035] Example 1

[0036] like Figure 1 As shown, in this embodiment, a regional air conditioning load prediction method based on a time series large model includes the following steps:

[0037] S1: Obtain historical air-conditioning load data and pre-process the regional air-conditioning load: Considering that the air-conditioning load is affected by the complex influence of indoor load disturbance and outdoor environmental factors, such as indoor equipment operation, personnel flow and outdoor weather data, the Spearman correlation analysis method is used to perform correlation analysis on the variables with high correlation with the current load to reduce excessive variables.

[0038] The Spearman correlation analysis method is used. This analysis method aims to solve the problem of information redundancy that may be caused by too many variables. Through the Spearman correlation analysis, the key variables that are highly correlated with the current load are screened out, thereby reducing the calculation complexity and improving the prediction efficiency while ensuring the prediction accuracy. The specific formula is:

[0039]

[0040] In the formula, is the rank difference, and n is the number of data.

[0041] LOWESS is used to reduce the noise of load data with complex structures. In the actual operation environment, air conditioning load data is often interfered by multiple factors, resulting in certain noise in the data. In order to improve data quality and ensure the accuracy of prediction, it is particularly important to reduce the noise of air conditioning load data. In addition, since air conditioning operation load data usually exhibits nonlinear characteristics, traditional linear fitting methods may not be able to fully capture its complex change patterns. Specifically, LOWESS is used to smooth the air conditioning load data, which is specifically: a non-parametric learning method for data denoising, which is particularly suitable for data sets with complex structures. This method is implemented by the LOWESS (local weighted scatter plot smoothing) algorithm to avoid overfitting and improve the noise reduction effect. The LOWESS algorithm uses polynomial weighted fitting, which is mainly based on local observation data, and the error estimation is performed by the least squares method. Specifically, the weight function in the LOWESS algorithm is selected as a tricubic weight function, which can enhance the chi-square distribution approximation of the error variance estimation, and can provide sufficient smoothness in many cases, ensuring the noise reduction effect while maintaining the authenticity of the data, laying a solid foundation for subsequent short-term prediction based on large time series models.

[0042] The specific form of the tricubic weight function is shown in formula (2). The selection of this function aims to further improve the applicability and effect of the LOWESS algorithm in data denoising.

[0043]

[0044] Where x={x1,x2,…,x n}, x is the distance from a given data point to a point on the fitted curve, scaled to the range of 0 to 1, and n is the number of sampling points.

[0045] Weighted Least Squares Regression:

[0046]

[0047] In the formula, ω For sample X The weight of .

[0048] Smoothing Data

[0049]

[0050] The width of the window p affects the noise reduction effect, which is expressed by formula (5):

[0051] p=τ*n (5)

[0052] Where n is the number of data sampling points and τ is the proportional coefficient.

[0053] S2: Process the air conditioning load data and key variables through the prompt technology to obtain the processed data sequence;

[0054] S3: Build a time series prediction model based on the UniTS model; train and test the model through the processed data sequence to obtain a trained time series prediction model;

[0055] The model converts the original time series data into three types of tokens, namely sequence token, prompt token and task token. Sequence token divides the original sequence into multiple patch fragments through non-overlapping windows. Each patch fragment is mapped by a multi-layer perceptron (MLP) to obtain the corresponding embedding to characterize the sequence characteristics of the original time series data; prompt token is a learnable embedding used to absorb contextual information in the sequence according to the characteristics of the task, and each task has its own unique prompt embedding; task token includes mask token and cls token. Mask token is used to uniformly process time series prediction and filling tasks, and cls token is used for time series classification tasks, thereby realizing the unified processing of different types of time series tasks.

[0056] The UniTS time series large model is specifically as follows: the structure consists of N repeatedly stacked UNITS blocks and a classification layer (CLS). Each UNITS block contains sequence multi-head self-attention (MHSA), variable MHSA, dynamic multi-layer perceptron (MLP) and gate modules, which are designed to process data from multiple domains and adapt to different numbers of variables and sequence points. These components work together to efficiently process multivariate data sequences from different fields. The model first receives the data sequence processed by Prompt, and then processes it through the UNITS block to extract key information. During the processing, task-related instructions are passed to the mask and CLS layers to realize the prediction conversion of data sequences to generation tasks or recognition tasks, and convert data sequences into prediction results of generation tasks or recognition tasks, thereby realizing efficient integration and intelligent analysis of cross-domain data.

[0057] The specific principle of sequence multi-head self-attention (MHSA) is as follows: The sequence multi-head self-attention (MHSA) mechanism is suitable for processing time series data. The mechanism divides the time series data into patch sequences and applies self-attention technology on these patch sequences. In addition, in order to capture the correlation information between variable dimensions, the mechanism performs pooling operations on the query and key in the attention on the time-wise (time dimension), and then performs overall self-attention on the variable dimension, further enhancing the model's ability to model dense relationships between different sequence lengths, so as to achieve a comprehensive understanding and effective modeling of time series data;

[0058]

[0059] where mean l is the average along the sequence dimension, and then we get:

[0060]

[0061] in is the attention map between variables, which is shared for all sequence points. For simplicity, the notation of multi-head attention is omitted.

[0062] Dynamic MLP: In order to solve the problem that different data input lengths lead to inconsistent input lengths of the MLP network followed by the traditional attention mechanism, which in turn causes changes in the MLP weight size, a dynamic linear operator (Dylinear) is introduced. The introduced dynamic linear operator (Dylinear) is specifically: It is simple and effective for modeling dense relationships between different sequence lengths. The operator is specially designed to adapt to sequences of different lengths in different data sources. It is centered on a weight interpolation scheme to adapt to these changing sequence lengths. Specifically, for a given sequence label Zs, the length is ls, and the predefined weights The Dylinear operation is as follows:

[0063] DyLinear(z s ; w) = W interp z s ; W interp =Interp(w) (11)

[0064] where Interp is a bilinear interpolation used to resize w from shape wi×wo to match the input sequence and the expected output length ls×lout.

[0065] This method uses bilinear interpolation technology to dynamically adjust the original MLP weights according to the actual length of the input vector, thereby generating MLP weights that exactly match the input dimension;

[0066]

[0067] where DyLinearM processes the sequence using two DyLinear operators and prompts to kens in the , and skip the CLS token to ensure consistency across all tasks. and This route separation leads to a scale combination effect and enhances multi-scale processing capabilities.

[0068] The Gray Wolf Optimization Algorithm is selected to focus on the optimization of network parameters; the Gray Wolf Optimization Algorithm (GWO) has the following principle: In the optimization process of the GWO (Grey Wolf Optimization) algorithm, the selection of individual fitness function is crucial to distinguish the quality of individuals in the population. In order to ensure the rationality of the fitness function, it is usually set as the objective function in practical applications. The quality of individuals is evaluated by comparing the size of the objective function value, and the level of individuals in the population is divided accordingly. The optimization goal of GWO is to make the difference between the predicted value of the air-conditioning load output by the network and the actual value as small as possible, that is, the objective function is shown in formula (6), where y' wt is the actual value of air conditioning load, y wt is the predicted value, L w is the length of the time series.

[0069]

[0070] This paper uses the mean square error (MSE) as an evaluation index of the degree of similarity between the actual value and the predicted value (i.e., the accuracy of the predicted value). The fitness function of the GWO algorithm set in this paper is shown in formula (7), where is the number of iterations. When ftness(t) is the smallest, the MSE of the prediction model is in the optimal state. When the network structure is set using this hyperparameter, the model has the optimal prediction accuracy.

[0071] fitness(t w )=(y' wt -y wt ) T (y' wt -y wt ) (7)

[0072] The steps of gray wolf algorithm optimization are as follows: initialize network parameters, clearly set the value range of hidden layers, number of blocks, number of iterations and step size; secondly, initialize the GWO algorithm, including the size of the gray wolf group and the initial position of each gray wolf individual, as well as the number of iterations and parameter settings of the algorithm, the position of each gray wolf, the number of iterations and parameter A W ,C W ,a w ; Next, calculate the fitness function value of the gray wolf individuals, sort the gray wolf individuals according to their fitness, and mark the top three gray wolf individuals with the best fitness as w α ,w β ,w δ ; Calculate the remaining gray wolf individuals ω and w in the population according to formula (8) α ,w β ,w δ distance, and update the position of the gray wolf ω according to formula (8); if the GWO algorithm reaches the preset maximum number of iterations or the optimization result meets the predetermined accuracy requirements, the algorithm ends and outputs the optimal network parameters; otherwise, repeat the steps of fitness calculation, sorting and gray wolf position update; finally, the optimized network parameters are applied to the network model.

[0073]

[0074] Among them, W(t w +1) for Gray Wolf W iw At the position of the tw+1th iteration, W′1(t w )、W′2(t w )、W′3(t w ) are gray wolf W iw The vector moving toward α wolf, β wolf, and δ wolf, r w2 is a random number in the interval [0,1], k = 1, 2, 3, a w is the convergence factor, a w =2-2(t w / t wmax ), t wmax is the maximum number of iterations. When the number of iterations increases, the convergence factor a decreases linearly to 0. wk It also changes accordingly, which may be greater than 1 or less than 1, and finally achieves the goal of controlling the optimization direction.

[0075] S4: Regional air conditioning load forecasting is performed using the trained time series forecasting model.

[0076] In this embodiment, the air conditioning load data, temperature, humidity, etc. of a certain area in August 2023 are analyzed. The prediction model of the present invention is compared with the widely used GRU neural network prediction model and the LightGBM prediction model. The mean absolute percentage error (MAPE) is selected as the evaluation criterion. This indicator can intuitively reflect the degree of deviation between the prediction model and the actual load data. Through comparative experiments, the air conditioning load prediction capabilities of various models in this specific area can be comprehensively evaluated, thereby providing a strong reference for future research and application of prediction methods.

[0077] Specifically, the relationship between temperature, humidity and air conditioning load in the region in August 2023 was used as an example for data analysis. In the process of statistically analyzing the load data of air conditioning daily load, the influence of factors such as holidays and rest days on the maximum daily load of air conditioning was fully considered. The correlation coefficient between temperature, humidity and load was calculated by applying the Spearman correlation coefficient. The results show that there is a close correlation between the meteorological factors and air conditioning load in the region. The correlation analysis is as follows Figure 2 shown.

[0078] In the initial stage, experience-based parameters were set for the GRU neural network prediction model and the LightGBM prediction model. For the GRU model, the number of hidden layer neurons was set to 10, the learning rate was 0.001, and the number of iterations was 100. For the LightGBM model, the number of weak regression trees was set to 100, the number of leaves was 30, the learning rate was 0.001, the L1 and L2 regularization parameters were 0.01 and 0.09 respectively, and the number of iterations was 3000. The relevant parameters of the time series large model remained in their initial state at this time without adjustment.

[0079] After multiple parameter adjustments, for the GRU model, the learning rate was adjusted to 0.004 to maintain higher learning efficiency and stability, while the number of iterations was maintained at 100. For the LightGBM model, the learning rate was adjusted to 0.002, and the L1 and L2 regularization parameters were set to 0.02 and 0.07, respectively, to optimize the generalization ability of the model. The number of weak regression trees and leaves remained unchanged, and the number of iterations remained 3000. The Gray Wolf Optimization Algorithm is used to obtain the hyperparameters of the time series large model. The air conditioning load is predicted by the GRU model, LightGBM model and the time series prediction large model in this embodiment after parameter adjustment, and the prediction error of the prediction model is calculated based on the prediction results. The results are shown in Table 1.

[0080] Table 1 Comparison of prediction errors of multiple prediction models

[0081] Model GRU LightGBM Time series large model Before optimization MAPE% 7.667 8.475 6.402 After optimization MAPE% 6.587 6.425 6.281

[0082] From Table 1, Figure 3 and Figure 4 It can be seen that the deviations of the prediction algorithms using the initial parameters are large, with the largest prediction error at 6 points. The average absolute percentage error using the time series large model is 6.402%. The average absolute percentage error of the optimized time series large model is 6.281%. The time series model can effectively reduce the impact of real-time meteorological factors on air conditioning load prediction and improve prediction accuracy.

[0083] Example 2

[0084] In this embodiment, a regional air conditioning load prediction system based on a time series large model includes:

[0085] A data acquisition module is used to obtain historical air conditioning load data and key variables related to air conditioning load;

[0086] The data processing module is used to process the air conditioning load data and key variables through the prompt technology to obtain the processed data sequence;

[0087] The model building module is used to build a large time series prediction model based on the UniTS model; the model is trained and tested by the processed data sequence to obtain a trained large time series prediction model;

[0088] The prediction module is used to predict regional air conditioning load through the trained time series prediction model.

[0089] Example 3

[0090] Those skilled in the art should understand that an embodiment of the present application may provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned method for regional air conditioning load prediction based on a large time series model when executing the computer program.

[0091] Example 4

[0092] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0093] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0094] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0095] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0096] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.

Claims

1. A regional air conditioning load forecasting method based on a time series large model, characterized in that: include: Obtain historical air conditioning load data and key variables related to air conditioning load; The air conditioning load data and key variables are processed by Prompt technology to obtain the processed data sequence; Build a large time series prediction model based on the UniTS model; Train and test the model through the processed data sequence to obtain a trained time series prediction model; Regional air conditioning load forecasting is performed using a trained time series forecasting model.

2. The regional air conditioning load prediction method based on the time series large model according to claim 1 is characterized in that: The step of obtaining key variables related to air conditioning load includes: Through Spearman correlation analysis, the key variables related to air conditioning load are screened out, and the Spearman correlation coefficient p calculation formula is: in, is the rank difference, and n is the number of data.

3. The regional air conditioning load prediction method based on the time series large model according to claim 1 is characterized in that: The step of obtaining historical air conditioning load data includes: preprocessing the obtained historical air conditioning load data, the preprocessing includes smoothing the historical air conditioning load data using a local weighted scatter plot smoothing algorithm; the local weighted scatter plot smoothing algorithm uses polynomial weighted fitting and performs error estimation through the least squares method; the local weighted scatter plot smoothing algorithm uses a tricubic weight function.

4. The regional air conditioning load prediction method based on the time series large model according to claim 1 is characterized in that: The large time series prediction model based on the UniTS model includes N stacked UNITS blocks and a CLS component, and the UNITS block includes a sequence multi-head self-attention MHSA, a variable MHSA, a dynamic multi-layer perceptron MLP and a gate module; the large time series prediction model based on the UniTS model converts the original sequence into three types of token tags, including sequence token, prompt token and task token, the sequence token divides the original sequence into multiple patch fragments through non-overlapping windows, and each patch fragment is mapped by a dynamic multi-layer perceptron MLP to obtain a corresponding embedding embedding to characterize the sequence characteristics of the original sequence; the prompt token is used to absorb the contextual information in the original sequence according to the task characteristics; the task token includes a mask token and a cls token, the mask token is used to uniformly process the time series prediction and filling tasks to generate prediction results, and the cls token is used for the time series classification task.

5. The regional air conditioning load prediction method based on the time series large model according to claim 4 is characterized in that: The method of training and testing the model through the processed data sequence to obtain a trained time series prediction large model includes: using the Gray Wolf Optimization Algorithm to optimize the network hyperparameters in the time series prediction large model with the goal of minimizing the difference between the air conditioning load prediction value output by the time series prediction large model and the actual value of the air conditioning load.

6. The regional air conditioning load prediction method based on the time series large model according to claim 4 is characterized in that: The dynamic multi-layer perceptron MLP uses a dynamic linear operator Dylinear to process the original MLP weights, performs bilinear interpolation according to the length of the input vector, and generates MLP weights that match the input dimension. The Dylinear operation is as follows: DyLinear(from s ;in)=in interp With s ;IN interp =Interp(in) Among them, Zs is a given sequence label, w is the weight, Interp is the bilinear interpolation.

7. A regional air conditioning load forecasting system based on a time series large model, characterized in that: include: A data acquisition module is used to obtain historical air conditioning load data and key variables related to air conditioning load; The data processing module is used to process the air conditioning load data and key variables through the prompt technology to obtain the processed data sequence; Model building module, used to build a large time series prediction model based on the UniTS model; Train and test the model through the processed data sequence to obtain a trained time series prediction model; The prediction module is used to predict regional air conditioning load through the trained time series prediction model.

8. The regional air conditioning load forecasting system based on the time series large model according to claim 7 is characterized in that: The data acquisition module is also used to screen out key variables related to air conditioning load through Spearman correlation analysis. The Spearman correlation coefficient p calculation formula is: in, is the rank difference, and n is the number of data.

9. The regional air conditioning load forecasting system based on the time series large model according to claim 7 is characterized in that: The data acquisition module is also used to preprocess the acquired historical air conditioning load data, and the preprocessing includes smoothing the historical air conditioning load data using a local weighted scatter plot smoothing algorithm; the local weighted scatter plot smoothing algorithm uses polynomial weighted fitting and performs error estimation through the least squares method; the local weighted scatter plot smoothing algorithm uses a tricubic weight function.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

11. 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 method according to any one of claims 1 to 6 are implemented.

Images