Accurate energy consumption load prediction model and method and readable storage medium

Abstract

The invention discloses a precise energy load prediction model and method and a readable storage medium, and belongs to the field of energy load prediction. The problem that the prediction precision of an existing energy load prediction model is difficult to meet the actual demand of fine regulation and control of the smart power grid is solved. According to the technical scheme, the precise energy consumption load prediction model comprises a data acquisition module, and the data acquisition module is sequentially in communication connection with a multi-source data fusion module, a precise user clustering module, a multi-dimensional feature mining module, an intelligent hybrid model construction module and a model optimization and adaptive adjustment module. The model optimization and adaptive adjustment module comprises a model optimization unit and an adaptive adjustment unit, the multi-source data fusion module comprises a data cleaning unit, and the data cleaning unit is sequentially in communication connection with a data standardization unit, a feature extraction unit and a data fusion unit; the method is applied to energy load prediction.

Description

Technical Field

[0001] This invention provides an accurate energy load forecasting model, method, and readable storage medium, belonging to the field of energy load forecasting technology. Background Technology

[0002] In the fields of energy management and power system operation, load forecasting is a core component for achieving efficient dispatch and optimal resource allocation. With energy demand growing exponentially and the energy structure rapidly evolving towards multi-energy complementarity, traditional load forecasting methods are facing unprecedented technological bottlenecks.

[0003] From a data perspective, the existing forecasting system has significant structural defects: the problem of a single data source is prominent, with an over-reliance on internal metering data of power companies. It lacks systematic integration of key external variables such as meteorological conditions (such as temperature and humidity), policy regulation (such as electricity price adjustments), and macroeconomic indicators (such as GDP growth rate), which makes the forecasting model unable to effectively capture the nonlinear variation of actual load.

[0004] From the perspective of user profile construction, traditional classification methods still rely on basic attributes such as voltage level and power capacity, which are difficult to adapt to the cyclical production power consumption characteristics of industrial users, the fluctuating characteristics of promotional activities of commercial users, and the seasonal differences in power consumption patterns of residential users, resulting in inaccurate predictions in specific scenarios.

[0005] From a technical implementation perspective, traditional prediction frameworks have two limitations: the data processing stage often uses simple statistical analysis and traditional feature engineering, which makes it difficult to break through the constraints of shallow data representation and cannot uncover the chaotic characteristics and spatiotemporal coupling relationships in the load sequence; the model building stage lacks an adaptive optimization mechanism, and when faced with complex scenarios such as the tiered load growth of industrial parks and sudden peak electricity consumption in commercial complexes, it is unable to dynamically adjust the model architecture and hyperparameters, making it difficult to meet the actual needs of the refined regulation of smart grids in terms of prediction accuracy. Summary of the Invention

[0006] To address the technical problem that the prediction accuracy of existing energy load forecasting models is insufficient to meet the actual needs of refined regulation in smart grids, this invention proposes an accurate energy load forecasting model, method, and readable storage medium.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: a precise energy load prediction model, including a data acquisition module, wherein the data acquisition module is communicatively connected to a multi-source data fusion module;

[0008] The precise user clustering module, which communicates with the output of the multi-source data fusion module, is used to cluster the fused historical load dataset and feature dataset to obtain multiple user classification feature data.

[0009] The multi-dimensional feature mining module, which communicates with the output of the precise user clustering module, obtains the stability features of user electricity consumption habits and load elasticity features based on historical load feature data and multiple user classification feature data. It then concatenates the stability features of user electricity consumption habits and load elasticity features with historical load feature data and multiple user classification feature data to form a multi-dimensional feature vector.

[0010] The intelligent hybrid model building module, which communicates with the output of the multi-dimensional feature mining module, is used to output third-party electricity consumption forecast data and analyze user load forecast values ​​for a future period of time.

[0011] The model optimization and adaptive adjustment module is connected to the intelligent hybrid module construction module. The model optimization and adaptive adjustment module includes a model optimization unit and an adaptive adjustment unit. The model optimization unit is used to optimize the parameters of the third electricity consumption prediction data. The adaptive adjustment unit is connected to the output of the multi-dimensional feature mining module and the output of the precise user clustering module, respectively, and is used to automatically adjust the structure of the intelligent hybrid model construction module and the parameters of the third electricity consumption prediction data.

[0012] Furthermore, the multi-source data fusion module includes a data cleaning unit, which is sequentially connected to a data standardization unit, a feature extraction unit, and a data fusion unit.

[0013] Furthermore, the data fusion unit includes a front-end CNN unit, which is sequentially connected to an attention mechanism unit and a bidirectional GRU unit.

[0014] Furthermore, the intelligent hybrid model building module includes Holt-Winters model units, which are communicatively connected to LSTM-Transformer model units.

[0015] A precise energy load forecasting method, based on the aforementioned precise energy load forecasting model, includes the following steps:

[0016] Step S1: Collect historical load datasets and feature datasets;

[0017] Step S2: The historical load dataset and feature dataset are sequentially processed through a data standardization unit, a feature extraction unit, and a data fusion unit to obtain the fused historical load dataset and feature dataset.

[0018] Step S3: Classify the fused historical load dataset and feature dataset from different dimensions to obtain preliminary user classification label data;

[0019] Step S4: Decompose the user preliminary classification label data and historical load dataset into different frequency scales using wavelet transform to obtain historical load feature data;

[0020] Step S5: Construct a clustering model based on the initial user classification label data, and obtain multiple user classification feature data through the k-means++ clustering algorithm to analyze the electricity consumption characteristics of each type of user.

[0021] Step S6: Based on historical load characteristic data and multiple user classification characteristic data, obtain user electricity consumption habit stability characteristics and load elasticity characteristics, and concatenate user electricity consumption habit stability characteristics, load elasticity characteristics, historical load characteristic data and multiple user classification characteristic data to form a multi-dimensional feature vector;

[0022] Step S7: Perform preliminary fitting on the fused historical load dataset and feature dataset to output the first electricity consumption prediction data. Then, horizontally concatenate the first electricity consumption prediction data and the multi-dimensional vector to form the second electricity consumption prediction data. Input the second electricity consumption prediction data into the LSTM-Transformer model unit for training and output the third electricity consumption prediction data to analyze the user load prediction value in the future period.

[0023] Step S8: After collecting the actual electricity load of users in the same time dimension using the model optimization and adaptive adjustment module and the third electricity prediction data output by the intelligent hybrid module, the parameters of the third electricity prediction data are optimized using the particle swarm optimization (PSO) algorithm.

[0024] Furthermore, the feature dataset in step S1 includes electricity consumption categories, contracted capacity, power generation methods, special user group data, industry category data, time data, meteorological data, policy data, and economic data.

[0025] Furthermore, in step S3, the fused historical load dataset and feature dataset are classified into five dimensions: electricity consumption category, contract capacity, power generation method, special user group and industry category to obtain user profile classification dataset. The user profile classification dataset is then labeled to output multiple user preliminary classification label data.

[0026] Furthermore, the historical load characteristic data in step S4 can reflect the trend of electricity load changes over time, seasonal changes, and fluctuations in electricity load within a cycle.

[0027] Furthermore, the calculation formula for the stability characteristics of user electricity consumption habits in step S6 is as follows:

[0028] ;

[0029] in, It is the standard deviation of historical load characteristic data. This represents the mean of historical load characteristic data;

[0030] The formula for calculating the load elasticity characteristic is:

[0031] ;

[0032] in For load elasticity, This represents the percentage change in historical load characteristic data. This represents the percentage change in electricity prices.

[0033] A readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described above.

[0034] The advantages of this invention over the prior art are as follows:

[0035] 1. The accurate energy load prediction model provided by this invention constructs a more scientific and comprehensive energy load prediction data support system by integrating historical load datasets and feature datasets. By clustering the integrated historical load datasets and feature datasets, multiple user classification feature data are obtained, realizing a refined classification of user electricity consumption behavior. On this basis, user electricity consumption habit stability features and load elasticity features are obtained. The user electricity consumption habit stability features, load elasticity features, historical load feature data, and multiple user classification feature data are concatenated to form a multi-dimensional feature vector, which can explore the coupling relationship between pairs of dimensions and further improve the predicted value of energy load.

[0036] 2. This invention employs a hybrid model combining LSTM-Transformer and Holt-Winters model units, and optimizes the parameters of the third-party electricity consumption prediction data using the Particle Swarm Optimization (PSO) algorithm. This enhances the model optimization and adaptive adjustment modules' ability to handle complex nonlinear data and their adaptive adjustment capabilities. This hybrid model better adapts to diverse energy consumption scenarios, improving the model's generalization ability and prediction accuracy. Attached Figure Description

[0037] The present invention will be further described below with reference to the accompanying drawings:

[0038] Figure 1 This is a schematic diagram of the structure of the present invention. Detailed Implementation

[0039] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate relative orientations or positional relationships and are used only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0040] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0041] like Figure 1 As shown, the present invention provides an accurate energy load prediction model, including a data acquisition module. The data acquisition unit is used to collect historical load datasets and feature datasets. The feature datasets include electricity consumption categories, contracted capacity, power generation methods, special user group data, industry category data, time data, meteorological data, policy data, economic data, etc.

[0042] The multi-source data fusion module, which is communicatively connected to the output of the data acquisition module, includes a data cleaning unit, which is sequentially communicatively connected to a data standardization unit, a feature extraction unit, and a data fusion unit.

[0043] The precise user clustering module, which communicates with the output of the multi-source data fusion module, is used to cluster the fused historical load dataset and feature dataset to obtain user classification feature data.

[0044] The multi-dimensional feature mining module, which communicates with the output of the precise user clustering module, obtains the stability features and load elasticity features of user electricity consumption habits based on historical load feature data and multiple user classification feature data. It then concatenates the stability features of user electricity consumption habits, the load elasticity features, historical load feature data, and multiple user classification feature data to form a multi-dimensional feature vector.

[0045] The intelligent hybrid model building module is connected to the output of the multidimensional feature mining module. The intelligent hybrid model building module includes a Holt-Winters model unit, which is connected to an LSTM-Transformer model unit.

[0046] A model optimization and adaptive adjustment module is connected to the intelligent hybrid module construction module. The model optimization and adaptive adjustment module includes a model optimization unit and an adaptive adjustment unit.

[0047] The model optimization unit is used to optimize the parameters of the third-party electricity consumption forecast data;

[0048] The adaptive adjustment unit is connected to the output of the multidimensional feature mining module and the output of the precise user clustering module, respectively, and is used to automatically adjust the structure of the intelligent hybrid model construction module and the parameters of the third electricity consumption prediction data.

[0049] The present invention provides a method for accurate energy load forecasting, based on the aforementioned accurate energy load forecasting model, comprising the following steps:

[0050] Step S1: Collect historical load datasets and feature datasets.

[0051] Step S2: The collected historical load dataset and feature dataset are sequentially processed through the data standardization unit, feature extraction unit, and data fusion unit to obtain the fused collected historical load dataset and feature dataset.

[0052] Specifically, the historical load dataset and feature dataset input to the data acquisition module are fed into a data cleaning unit that uses the isolated forest algorithm to detect outliers and remove noise interference such as missing values, outliers, and duplicate values ​​from the historical load dataset and feature dataset. Then, the cleaned historical load dataset and feature dataset are fed into a data standardization unit, which uses the Z-score standardization method to eliminate the dimensional differences between the historical load dataset and the feature dataset. The standardized historical load dataset and feature dataset are then fed into a feature extraction unit, where the Dynamic Time Warping (DTW) algorithm is used to align the standardized historical load dataset and feature dataset according to the time series, eliminating the impact of sampling frequency differences. Finally, the aligned historical load dataset and feature dataset are fed into a data fusion unit.

[0053] The data fusion unit includes a front-end CNN unit, which is sequentially connected to an attention mechanism unit and a bidirectional GRU unit.

[0054] The front-end CNN unit uses a three-dimensional convolutional kernel and alternates between 64 convolutional layers and max pooling layers. It automatically learns the spatial correlation between the historical load dataset and the feature dataset output by the feature extraction unit in the feature dimension and then inputs them into the bidirectional GRU unit.

[0055] The bidirectional GRU unit dynamically updates the aligned historical load dataset and feature dataset through collaborative computation of the reset gate and update gate, effectively mining the dependencies between the historical load dataset and feature dataset across multiple time scales, such as intra-week and intra-day. The specific mechanism is as follows:

[0056] 1. Regarding the gating and collaboration mechanism:

[0057] The reset gate (r) controls the degree of forgetting of the historical workload dataset. When the degree of forgetting is close to 0, the history is ignored and the focus is on the current input (capturing short-term changes). The update gate (z) controls the degree of memory update. When it is close to 0, it tends to favor short-term dependencies (intraday mode). When it is close to 1, it retains long-term dependencies (intraweek mode). Cooperative computing: through the nonlinear transformation of the gated sigmoid and tanh activation functions, the information flow is adaptively adjusted.

[0058] 2. Bidirectional capture of multi-scale dependencies:

[0059] The forward GRU in the bidirectional GRU unit captures short- to medium-term dependencies (such as intraday load fluctuations) through a past-to-future computation method (the forward GRU processes historical load datasets and feature datasets in chronological order). Then, an attention mechanism is employed to enhance the contribution of pre-defined key data in the historical load dataset and feature dataset to the final prediction data, achieving deep fusion of the historical load dataset and feature dataset in the spatiotemporal dimension. The deeply fused historical load dataset and feature dataset are then input into the backward GRU, which uses a future-to-past computation method—the reverse of the forward GRU's processing order—to capture long-term dependencies (such as weekly periodic patterns). Finally, the hidden states of the historical load dataset and feature dataset from both directions are concatenated or weighted to form a multi-scale feature representation.

[0060] 3. Multi-scale mining strategy:

[0061] The input combines the original time series data with derived features (daily / weekly statistics, time codes), and applies a stacked multi-layer bidirectional GRU + attention mechanism to enhance the ability to extract time series features.

[0062] Step S3: Classify the merged historical load dataset and feature dataset from different dimensions to obtain preliminary user classification label data.

[0063] Step S4: Decompose the user preliminary classification label data and historical load dataset into different frequency scales using wavelet transform to obtain historical load feature data.

[0064] Step S5: Construct a clustering model based on the initial user classification label data, and obtain multiple user classification feature data through the k-means++ clustering algorithm to analyze the electricity consumption characteristics of each type of user.

[0065] Specifically, the merged historical load dataset and feature dataset are classified according to multiple dimensions such as electricity consumption category, contract capacity, power generation method, special user groups, and industry category to obtain a user profile classification dataset. This user profile classification dataset is then labeled to output multiple preliminary user classification label data. In this embodiment, electricity consumption categories are divided into industrial electricity consumption, commercial electricity consumption, and residential electricity consumption, contract capacity can be divided according to different intervals, and industry categories cover various industries such as manufacturing and services.

[0066] Wavelet transform is used to decompose the initial user classification label data and historical load dataset into different frequency scales, obtaining historical load characteristic data that reflects the trend of electricity load changes over time, seasonal changes, and periodic fluctuations. This includes historical load characteristic data reflecting the trend of electricity load changes over time, seasonal changes, and periodic fluctuations at different time scales. The calculation formula for wavelet transform is: ,in These are wavelet transform coefficients. It is a historical workload dataset. It is a wavelet function, where a is the scaling parameter and b is the translation parameter.

[0067] A clustering model is constructed based on the initial user classification label data. Multiple user classification feature data are obtained through the k-means++ clustering algorithm to analyze the electricity consumption characteristics of each user category.

[0068] The clustering model is:

[0069] ,in, It is the objective function for clustering. It is an indicator variable, when data points Belongs to cluster center hour =1, otherwise c is 0, where c is the clustering result. It is a cluster center. is the initial classification label data for the i-th user, n is the total number of initial classification label data for users, and k is the number of clusters.

[0070] Step S6: Based on historical load characteristic data and multiple user classification characteristic data, obtain the stability characteristics and load elasticity characteristics of user electricity consumption habits. Concatenate the stability characteristics and load elasticity characteristics of user electricity consumption habits, and concatenate the historical load characteristic data and multiple user classification characteristic data to form a multi-dimensional feature vector, which can be used to analyze the coupling relationship between pairs of dimensions.

[0071] Specifically, based on historical load characteristic data, the coefficient of variation (CV) is used to quantify the stability of users' electricity consumption habits, resulting in the stability characteristics of users' electricity consumption habits. The calculation formula is as follows:

[0072] ;

[0073] in, It is the standard deviation of historical load characteristic data. The CV value is the average of historical load characteristic data. The smaller the CV value, the more stable the user's electricity consumption habits.

[0074] By simulating user classification feature data under different electricity prices, load elasticity features are extracted using the demand price elasticity formula, which is:

[0075] ;

[0076] in For load elasticity, This represents the percentage change in historical load characteristic data. This represents the percentage change in electricity prices.

[0077] Step S7: Perform preliminary fitting on the fused historical load dataset and feature dataset to output the first electricity consumption prediction data. Then, horizontally concatenate the first electricity consumption prediction data and the multi-dimensional vector to form the second electricity consumption prediction data. Input the second electricity consumption prediction data into the LSTM-Transformer model unit for training and output the third electricity consumption prediction data to analyze the user load prediction values ​​in the future.

[0078] The Holt-Winters model unit is used to initially fit the fused historical load dataset and feature dataset, capturing macroscopic features such as the time-varying trend, seasonal variation trend, and periodic fluctuation of electricity load in the historical load feature dataset. This leads to the construction of an electricity consumption prediction model, outputting the first electricity consumption prediction data. The Holt-Winters model unit includes three smoothing parameters: level parameters... Trend parameters and seasonal parameters For the t-th historical load characteristic data Its electricity consumption prediction model is as follows:

[0079] ;

[0080] in:

[0081] Level value update: ;

[0082] Trend value update: ;

[0083] Seasonal factor update: , where L is the length of the seasonal cycle.

[0084] The first electricity consumption forecast data and the multi-dimensional vector output by the multi-dimensional feature mining module are horizontally concatenated to form the second electricity consumption forecast data. The second electricity consumption forecast data is then input into the LSTM-Transformer model unit for training, and the third electricity consumption forecast data is output to analyze the user load forecast values ​​for a future period of time.

[0085] Among them, LSTM (Long Short-Term Memory) network can effectively handle the long-term dependency problem in the second electricity consumption prediction data. Its core structure includes input gates. Forgotten Gate Output gate and memory unit The calculation formula is as follows:

[0086] ;

[0087] ;

[0088] ;

[0089] , ;

[0090] in Here is the Sigmoid function, and ⊙ represents element-wise multiplication. For the bias vector, For input as the entry point for the current input The weight matrix The input gate hides the state from the previous time step. The weight matrix; Forget gate for the current input The weight matrix; To hide the previous state using the forget gate The weight matrix; For the memory cell candidate value for the current input The weight matrix; For candidate values ​​of memory cells, hide the state from the previous time step. The weight matrix; For the output gate to the current input The weight matrix; The output gate is set to the hidden state of the previous moment. The weight matrix.

[0091] The LSTM-Transformer model unit uses a self-attention mechanism to automatically focus on different locations in the input second electricity consumption prediction data, better capturing the complex nonlinear relationships in the second electricity consumption prediction data. The calculation process of the self-attention mechanism is as follows:

[0092] ;

[0093] ;

[0094] ;

[0095] ;

[0096] in The second electricity consumption forecast data was linearly transformed. The resulting query vector , They are the key vector and the value vector, respectively. For query vector The weight matrix is ​​used to calculate the query representation in the self-attention mechanism; Key vector The weight matrix is ​​used to calculate the key representation in the self-attention mechanism; Value vector The weight matrix is ​​used to calculate the value representation in the self-attention mechanism; Represents the key vector matrix Transpose of; Represents the query vector With the transposed key vector Perform matrix multiplication to obtain an attention score matrix; Scaling of attention score, where It represents the dimension of the key vector, and its purpose is to stabilize gradient propagation; The scaled attention score is converted into a probability distribution to represent the correlation between different locations; It is the output vector after weighted summation, which retains important information from the input second electricity consumption prediction data.

[0097] Step S8: After collecting the actual electricity load of users in the same time dimension using the model optimization and adaptive adjustment module and the third electricity prediction data output by the intelligent hybrid module, the parameters of the third electricity prediction data are optimized using the particle swarm optimization (PSO) algorithm.

[0098] Specifically, the mean square error (MSE) between the predicted electricity consumption data and the actual electricity load of users is calculated by constructing a fitness function. The smaller the MSE, the higher the fitness. The fitness function formula is as follows:

[0099] ;

[0100] in This is the actual load. It is a model that predicts the load.

[0101] The mean squared error (MSE) is fed back to the particle swarm optimization (PSO) algorithm to update the particle state. The particles dynamically adjust their velocity and position based on their own historical best position and the swarm's historical best position. By continuously iterating and updating the parameters of the intelligent hybrid model building module, the prediction accuracy of the intelligent hybrid model building module is improved.

[0102] The adaptive adjustment unit is connected to the output of the multi-dimensional feature mining module and the output of the precise user clustering module. It has an adaptive mechanism based on the user electricity consumption habit stability characteristics and load elasticity characteristics output by the multi-dimensional feature mining module, as well as historical load characteristic data, time characteristics, and feature datasets. It automatically adjusts the structure of the intelligent hybrid model construction module and the parameters of the third electricity consumption prediction data according to the energy consumption characteristics and trends of different industries and users.

[0103] The user classification feature data output by the precise user clustering module is divided into two types based on the stability characteristics of users' electricity consumption habits: fluctuating electricity users and stable electricity users. For fluctuating electricity users, the number of neurons in the hidden layers of the LSTM-Transformer model unit is increased to enhance the LSTM-Transformer model unit's ability to capture complex nonlinear relationships in the second electricity consumption prediction data. For stable electricity users, the number of neurons in the hidden layers of the LSTM-Transformer model unit is appropriately reduced to avoid overfitting.

[0104] The present invention provides a readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method described above.

[0105] Regarding the specific structure of this invention, it should be noted that the connection relationships between the various component modules used in this invention are definite and achievable. Except as specifically described in the embodiments, their specific connection relationships can bring about corresponding technical effects and solve the technical problems proposed by this invention without relying on the execution of corresponding software programs. The models of the components, modules, and specific components appearing in this invention, the connection methods between them, and the conventional usage methods and expected technical effects brought about by the above technical features, unless specifically described, are all publicly disclosed content in patents, journal articles, technical manuals, technical dictionaries, and textbooks that can be obtained by those skilled in the art before the application date, or belong to conventional technology, common knowledge, and other existing technologies in this field. There is no need to elaborate, which makes the technical solution provided in this case clear, complete, and achievable, and can reproduce or obtain corresponding physical products based on this technical means.

[0106] Finally, 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 foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A precise energy load prediction model, characterized in that, It includes a data acquisition module, which is connected to a multi-source data fusion module. The precise user clustering module, which communicates with the output of the multi-source data fusion module, is used to cluster the fused historical load dataset and feature dataset to obtain multiple user classification feature data. The multi-dimensional feature mining module, which communicates with the output of the precise user clustering module, obtains the stability features of user electricity consumption habits and load elasticity features based on historical load feature data and multiple user classification feature data. It then concatenates the stability features of user electricity consumption habits and load elasticity features with historical load feature data and multiple user classification feature data to form a multi-dimensional feature vector. The intelligent hybrid model building module, which communicates with the output of the multi-dimensional feature mining module, is used to output third-party electricity consumption forecast data and analyze user load forecast values ​​for a future period of time. The model optimization and adaptive adjustment module is connected to the intelligent hybrid module construction module. The model optimization and adaptive adjustment module includes a model optimization unit and an adaptive adjustment unit. The model optimization unit is used to optimize the parameters of the third electricity consumption prediction data. The adaptive adjustment unit is connected to the output of the multi-dimensional feature mining module and the output of the precise user clustering module, respectively, and is used to automatically adjust the structure of the intelligent hybrid model construction module and the parameters of the third electricity consumption prediction data.

2. The precise energy load prediction model according to claim 1, characterized in that, The multi-source data fusion module includes a data cleaning unit, which is sequentially connected to a data standardization unit, a feature extraction unit, and a data fusion unit.

3. The precise energy load prediction model according to claim 1, characterized in that, The data fusion unit includes a front-end CNN unit, which is sequentially connected to an attention mechanism unit and a bidirectional GRU unit.

4. The precise energy load prediction model according to claim 1, characterized in that, The intelligent hybrid model building module includes Holt-Winters model units, which are communicatively connected to LSTM-Transformer model units.

5. A method for accurate energy load forecasting, characterized in that, Based on the accurate energy load prediction model as described in any one of claims 1-4, the following steps are included: Step S1: Collect historical load datasets and feature datasets; Step S2: The historical load dataset and feature dataset are sequentially processed through a data standardization unit, a feature extraction unit, and a data fusion unit to obtain the fused historical load dataset and feature dataset. Step S3: Classify the fused historical load dataset and feature dataset from different dimensions to obtain preliminary user classification label data; Step S4: Decompose the user preliminary classification label data and historical load dataset into different frequency scales using wavelet transform to obtain historical load feature data; Step S5: Construct a clustering model based on the initial user classification label data, and obtain multiple user classification feature data through the k-means++ clustering algorithm to analyze the electricity consumption characteristics of each type of user. Step S6: Based on historical load characteristic data and multiple user classification characteristic data, obtain user electricity consumption habit stability characteristics and load elasticity characteristics, and concatenate user electricity consumption habit stability characteristics, load elasticity characteristics, historical load characteristic data and multiple user classification characteristic data to form a multi-dimensional feature vector; Step S7: Perform preliminary fitting on the fused historical load dataset and feature dataset to output the first electricity consumption prediction data. Then, horizontally concatenate the first electricity consumption prediction data and the multi-dimensional vector to form the second electricity consumption prediction data. Input the second electricity consumption prediction data into the LSTM-Transformer model unit for training and output the third electricity consumption prediction data to analyze the user load prediction value in the future period. Step S8: After collecting the actual electricity load of users in the same time dimension using the model optimization and adaptive adjustment module and the third electricity prediction data output by the intelligent hybrid module, the parameters of the third electricity prediction data are optimized using the particle swarm optimization (PSO) algorithm.

6. The accurate energy load forecasting method according to claim 5, characterized in that, The feature dataset in step S1 includes electricity consumption categories, contracted capacity, power generation methods, special user group data, industry category data, time data, meteorological data, policy data, and economic data.

7. The accurate energy load forecasting method according to claim 5, characterized in that, In step S3, the merged historical load dataset and feature dataset are classified into five dimensions: electricity consumption category, contract capacity, power generation method, special user group and industry category to obtain user profile classification dataset. The user profile classification dataset is then labeled to output multiple user preliminary classification label data.

8. The accurate energy load forecasting method according to claim 5, characterized in that, The historical load characteristic data in step S4 can reflect the trend of electricity load changes over time, seasonal changes, and fluctuations in electricity load within a cycle.

9. The accurate energy load forecasting method according to claim 5, characterized in that, The formula for calculating the stability characteristics of user electricity consumption habits in step S6 is as follows: ； in, It is the standard deviation of historical load characteristic data. This represents the mean of historical load characteristic data; The formula for calculating the load elasticity characteristic is: ； in For load elasticity, This represents the percentage change in historical load characteristics. This represents the percentage change in electricity prices.

10. A readable storage medium, characterized in that: The readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method as described in any one of claims 5 to 9.

Images