Power load prediction method and system

Abstract

The invention discloses a power load prediction method and system, and the method comprises the steps: obtaining multi-element load data, and carrying out the comprehensive correlation analysis of the multi-element load data, so as to obtain a high-correlation feature; the comprehensive correlation analysis comprises correlation analysis between meteorological factors and multi-element loads, coupling analysis between the multi-element loads and respective correlation analysis of the multi-element loads; constructing a hybrid deep learning prediction model through fusion of a convolutional neural network, a bidirectional long-short-term memory network and an attention mechanism; and inputting the high correlation feature into a mixed deep learning prediction model for prediction to obtain a predicted multi-element load value. According to the method, the dimensionality of the input layer of the multi-element load prediction model is reduced, and the input data quality is improved. The over-fitting phenomenon of a prediction algorithm can be effectively prevented, and the model prediction precision is improved; and the scheduling requirement of a system in actual engineering is met.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid load forecasting, and in particular to a power load forecasting method and system. Background Art

[0002] Load forecasting is a vital part of the integrated energy system and is crucial to the planning, operation and management of the power system. Accurate load forecasting is of great significance to the safe and stable operation of the system, resource optimization scheduling and market transactions. It can help system operators plan and prepare in advance to ensure the balance of energy supply and demand and avoid energy shortages or surpluses.

[0003] However, traditional forecasting methods such as time series analysis and regression analysis have limitations. The former assumes that the load data is linear and stable, and it is difficult to deal with nonlinear factors; the latter relies on artificial feature engineering, which is easy to introduce errors. Specifically, many statistical models in linear assumptions and nonlinear factors assume that the data is linear and obeys a certain probability distribution, which is not flexible enough when dealing with nonlinear factors such as temperature fluctuations and holiday effects, and it is difficult to capture complex patterns in load demand. For feature selection and engineering, methods such as regression analysis require manual selection of input variables and feature engineering, which is not only time-consuming and labor-intensive, but also prone to prediction bias due to improper feature selection. As for adaptability and generalization ability, with the continuous development of the power market and changes in user behavior, load patterns will also change. Traditional models often lack sufficient adaptability and generalization ability to cope with these changes. Therefore, there is an urgent need for a method to improve the accuracy and efficiency of multivariate load data prediction. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method and system for power load forecasting to solve the problems that the current existing technology is difficult to deal with, such as reliance on artificial feature engineering, easy introduction of errors, large load fluctuations, numerous nonlinear influencing factors, and difficulty in constructing data sets, which make load forecasting difficult, resulting in low prediction accuracy and low efficiency.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a method for predicting power load, comprising:

[0008] Acquire multivariate load data, and perform comprehensive correlation analysis on the multivariate load data to obtain high correlation characteristics; the comprehensive correlation analysis includes correlation analysis between meteorological factors and multivariate loads, coupling analysis between multivariate loads, and correlation analysis of each multivariate load;

[0009] A hybrid deep learning prediction model is constructed by integrating convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms;

[0010] The highly correlated features are input into a hybrid deep learning prediction model for prediction to obtain predicted multivariate load values.

[0011] As a preferred solution of the power load forecasting method described in the present invention, the correlation analysis between meteorological factors and multivariate loads is performed through the maximum mutual information coefficient, which is expressed as:

[0012]

[0013] Among them, MIC(X,Y) represents the MIC value of features X and Y; p(x) and p(y) represent the marginal probability distribution functions of features X and Y respectively; p(x,y) represents the joint probability distribution function of features X and Y; a and b represent the number of grids divided in the x and y directions in the two-dimensional space; B is a constant.

[0014] As a preferred solution of the power load forecasting method of the present invention, the coupling analysis between multiple loads is performed by using the Spearman correlation coefficient, which is expressed as:

[0015]

[0016] Among them, ρ x,y represents the correlation between the load and the input feature, ρ x,y ≤1, the closer to 1, the higher the correlation; x i ,y i is the value of the two factors of the ith data point; is called the mean of the two factors; n is the number of data points.

[0017] As a preferred solution of the power load forecasting method described in the present invention, the correlation analysis of multiple loads is performed by using an autocorrelation function, which is expressed as:

[0018]

[0019] Among them, k, ρ k represent the lag order and the ACF value of lag k respectively; t , O t-k They represent the load at time t and time tk respectively; Represents the average value of all loads.

[0020] As a preferred solution of the power load forecasting method described in the present invention, wherein: the high correlation features are input into the hybrid deep learning forecasting model for forecasting, including:

[0021] The convolutional neural network layer extracts spatial features from the input high-correlation feature data to obtain a spatial feature set;

[0022] The spatial feature set is input into the BiLSTM layer. The BiLSTM layer trains multiple LSTM units from the forward and backward directions respectively. The forward and backward information features of the data before a certain moment of the spatial feature set are read simultaneously through the hidden layer to obtain the relationship between the current moment data and the past and future moment data. The current moment data is fitted from two directions to obtain the power load data at a certain moment in the future.

[0023] As a preferred solution of the power load forecasting method described in the present invention, wherein: the high correlation features are input into the hybrid deep learning forecasting model for forecasting, further comprising:

[0024] The attention mechanism layer assigns different weights to each feature, and the weights are used to calculate the attention score through the softmax function;

[0025] The weighted features are input into the fully connected layer for prediction, and the predicted multivariate load value is output.

[0026] As a preferred solution of the power load forecasting method described in the present invention, wherein: the attention score is obtained by calculating the similarity between the query vector and each input vector, and there are n attention scores represented as: s1, s2, s3, ..., s n ;

[0027] The Softmax function converts the original attention score into a weight in the form of a probability distribution. For any attention score s i , and its weight calculation formula after softmax is:

[0028]

[0029] in, Represents the attention score s i Take the index, is the sum of all attention scores after exponentiation;

[0030] The calculated weights are used to perform weighted summation on the corresponding input elements. The output vector after attention weighting is expressed as {v1,v2,…,v n}, the output is: Among them, ω i is the corresponding weight.

[0031] In a second aspect, the present invention provides a power load forecasting system, comprising:

[0032] The acquisition module is used to acquire multivariate load data and perform comprehensive correlation analysis on the multivariate load data to obtain high correlation characteristics; the comprehensive correlation analysis includes correlation analysis between meteorological factors and multivariate loads, coupling analysis between multivariate loads, and correlation analysis of each multivariate load.

[0033] Building blocks for constructing hybrid deep learning prediction models by integrating convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms;

[0034] The prediction module is used to input the high correlation features into the hybrid deep learning prediction model for prediction to obtain the predicted multivariate load value.

[0035] In a third aspect, the present invention provides a computing device, comprising:

[0036] Memory and processor;

[0037] The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the power load forecasting method are implemented.

[0038] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the power load forecasting method.

[0039] Compared with the prior art, the present invention has the following beneficial effects: the present invention uses data correlation analysis technologies such as MIC, PCC, and ACF to integrate into a feature comprehensive correlation analysis method, performs correlation analysis between meteorological factors and loads, between loads and load coupling, and before and after loads, proposes high correlation features, reduces the dimension of the input layer of the multivariate load forecasting model, and improves the quality of input data. The organic combination of CNN, BiLSTM, and Attention is used to predict data. While fully utilizing the advantages of CNN in effectively extracting spatial features and BiLSTM network in extracting bidirectional temporal features of sequence data, the Attention mechanism is better utilized to effectively reduce the loss of historical information and highlight the information of key historical time points to reduce the impact of redundant information on load forecasting results. This can effectively prevent the overfitting phenomenon of the prediction algorithm and improve the prediction accuracy of the model. More in line with the scheduling requirements of the system in actual engineering, this prediction method can provide good support for the system's day-ahead scheduling and intraday operation optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0041] Figure 1 This is a schematic diagram of the overall process logic of the power load forecasting method according to the first embodiment of the present invention;

[0042] Figure 2 This is a prediction effect diagram before the electric load model is integrated in the electric load prediction method described in the second embodiment of the present invention;

[0043] Figure 3 This is a prediction effect diagram after the electric load model is integrated in the electric load prediction method described in the second embodiment of the present invention;

[0044] Figure 4 A prediction error diagram before the electric load model is integrated in the electric load prediction method according to the second embodiment of the present invention;

[0045] Figure 5 This is a prediction error diagram after the electric load model is integrated in the electric load prediction method described in the second embodiment of the present invention. DETAILED DESCRIPTION

[0046] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0047] Example 1

[0048] Reference Figure 1 , as an embodiment of the present invention, provides a method for predicting power load, comprising:

[0049] S100: acquiring multivariate load data, and performing comprehensive correlation analysis on the multivariate load data to obtain high correlation characteristics; the comprehensive correlation analysis includes correlation analysis between meteorological factors and multivariate loads, coupling analysis between multivariate loads, and correlation analysis of each multivariate load;

[0050] In the embodiment of the present application, the correlation analysis between the meteorological factors and the multivariate loads is performed by using the maximum mutual information coefficient, which is expressed as:

[0051]

[0052] Among them, MIC(X,Y) represents the MIC value of features X and Y; p(x) and p(y) represent the marginal probability distribution functions of features X and Y respectively; p(x,y) represents the joint probability distribution function of features X and Y; a and b represent the number of grids divided in the x and y directions in the two-dimensional space; B is a constant.

[0053] Specifically, B is generally approximately 0.6 power of the amount of data.

[0054] It should be noted that the maximum mutual information coefficient is an evaluation indicator that can deeply mine the correlation between variables in big data. In the feature screening process, the maximum normative information under different interval divisions is used to deeply mine the correlation between input features and output features, thereby achieving the effect of accurately measuring the correlation between feature sequences and photovoltaic sequences.

[0055] In the embodiment of the present application, the coupling analysis between the multivariate loads is performed by using the Spearman correlation coefficient, which is expressed as:

[0056]

[0057] Among them, ρ x,y represents the correlation between the load and the input feature, ρ x,y ≤1, the closer to 1, the higher the correlation; x i ,y i is the value of the two factors of the ith data point; is called the mean of the two factors; n is the number of data points.

[0058] In the embodiment of the present application, the correlation analysis of each multivariate load is performed by using the autocorrelation function, which is expressed as:

[0059]

[0060] Among them, k, ρ k represent the lag order and the ACF value of lag k respectively; t , O t-k They represent the load at time t and time tk respectively; Represents the average value of all loads.

[0061] It should be noted that multivariate load forecasting requires multiple loads to be used as input variables and forecasted at the same time, and different loads have different characteristics at different time scales. Coupled with the variability of meteorological factors, this will cause the input of the forecasting model to increase sharply, which will increase the calculation amount of the model. In view of the above problems, a single feature correlation analysis method can no longer meet the requirements of forecasting accuracy. Therefore, the above scheme is used to conduct a comprehensive analysis of the features to simplify the input of the model and improve the forecasting accuracy of the model.

[0062] S200: Constructing a hybrid deep learning prediction model by integrating convolutional neural networks, bidirectional long short-term memory networks and attention mechanisms;

[0063] It should be noted that the convolutional neural network (CNN) is a deep learning model with efficient feature extraction capabilities. It consists of two core layers: a convolutional layer and a pooling layer. The convolutional layer effectively extracts the nonlinear local features of multivariate loads through convolution operations and outputs feature maps; the pooling layer is used to compress the size of the feature map, that is, to reduce the dimensionality of the input data while retaining important features and improving the generalization ability of the features.

[0064] Among them, the long short-term memory network (LSTM) is a recurrent neural network structure, and the bidirectional long short-term memory network (BiLSTM) is a further extension of it, consisting of LSTM in both forward and backward directions.

[0065] Among them, the attention mechanism can accurately capture the importance of input features, that is, assign different weights to different features, assign larger weights to key factors, and assign smaller weights to weakly related features to distinguish the importance of input features, thereby improving the model's processing efficiency for input features.

[0066] The organic combination of CNN, BiLSTM, and Attention is used to predict data. While fully leveraging the advantages of CNN in effectively extracting spatial features and BiLSTM network in extracting bidirectional temporal features of sequence data, the Attention mechanism is better used to effectively reduce the loss of historical information and highlight the information of key historical time points to reduce the impact of redundant information on load forecasting results. This can effectively prevent the overfitting phenomenon of the prediction algorithm and improve the prediction accuracy of the model.

[0067] S300: Inputting the high correlation features into the hybrid deep learning prediction model for prediction to obtain predicted multivariate load values.

[0068] In an embodiment of the present application, the high correlation features are input into the hybrid deep learning prediction model for prediction, including:

[0069] The convolutional neural network layer extracts spatial features from the input high-correlation feature data to obtain a spatial feature set;

[0070] The spatial feature set is input into the BiLSTM layer. The BiLSTM layer trains multiple LSTM units from the forward and backward directions respectively. The forward and backward information features of the data before a certain moment of the spatial feature set are read simultaneously through the hidden layer to obtain the relationship between the current moment data and the past and future moment data. The current moment data is fitted from two directions to obtain the power load data at a certain moment in the future.

[0071] In an embodiment of the present application, the high correlation features are input into the hybrid deep learning prediction model for prediction, which also includes:

[0072] The attention mechanism layer assigns different weights to each feature, and the weights are used to calculate the attention score through the softmax function;

[0073] The weighted features are input into the fully connected layer for prediction, and the predicted multivariate load value is output.

[0074] In the embodiment of the present application, the attention score is obtained by calculating the similarity between the query vector and each input vector, and there are n attention scores represented as: s1, s2, s3, ..., s n ;

[0075] The Softmax function converts the original attention score into a weight in the form of a probability distribution. For any attention score s i , and its weight calculation formula after softmax is:

[0076]

[0077] in, Represents the attention score s i Take the index, is the sum of all attention scores after exponentiation;

[0078] The calculated weights are used to perform weighted summation on the corresponding input elements. The output vector after attention weighting is expressed as {v1,v2,…,v n}, the output is: Among them, ω i is the corresponding weight.

[0079] The above is a schematic scheme of a power load forecasting method of this embodiment. It should be noted that the technical scheme of the power load forecasting system and the technical scheme of the power load forecasting method described above belong to the same concept, and the details of the technical scheme of the power load forecasting system in this embodiment that are not described in detail can all be referred to the description of the technical scheme of the power load forecasting method described above.

[0080] The power load forecasting system in this embodiment includes:

[0081] The acquisition module is used to obtain multivariate load data and conduct comprehensive correlation analysis on the multivariate load data to obtain high correlation characteristics; the comprehensive correlation analysis includes the correlation analysis between meteorological factors and multivariate loads, the coupling analysis between multivariate loads, and the correlation analysis of each multivariate load.

[0082] Building blocks for constructing hybrid deep learning prediction models by integrating convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms;

[0083] The prediction module is used to input the highly correlated features into the hybrid deep learning prediction model for prediction to obtain the predicted multivariate load value.

[0084] This embodiment further provides a computing device, which is applicable to power load forecasting, and includes:

[0085] Memory and processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the power load forecasting method proposed in the above embodiment.

[0086] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for realizing power load prediction as proposed in the above embodiment is implemented.

[0087] The storage medium proposed in this embodiment and the method for realizing power load prediction proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0088] Through the above description of the implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course can also be implemented by hardware. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ReadOnly, Memory, ROM), random access memory (Random Access Memory, RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform the methods of various embodiments of the present invention.

[0089] Example 2

[0090] Refer to Table 1-Table 2 and Figure 2-Figure 5Based on the previous embodiment, this embodiment provides an application case of a power load forecasting method to illustrate the feasibility and beneficial effects of our solution.

[0091] In order to verify the performance of the prediction model proposed in the present invention, an actual data set is used for experimental test analysis. The data comes from the measured data of cold, hot, electric and gas loads from January 1, 2020 to December 30, 2020 provided by a comprehensive energy system in a certain place in southern my country, with a sampling frequency of 1h and 24 sampling points per day. Each set of raw data contains 13 external environmental factors including month, hour, day, holiday type, wind speed, ambient temperature, air pressure, radiation, rainfall, dry bulb temperature, dew point temperature, wet bulb temperature and humidity, as well as load power data, totaling 8760 sets of data (6130 sets for training and 2628 sets for testing).

[0092] The mutual correlation between multivariate load and meteorological factors in the meteorological factor correlation results is different. The larger the MIC value, the greater the correlation between the feature and the output. According to the empirical method, the MIC threshold of the feature and the electric load is set to 0.5, and the MIC value of the ambient temperature and the electric load is 0.5172, then the ambient temperature is selected as the input of the electric load prediction data set; if the heat load MIC>0.5 is set, the ambient temperature and air pressure are selected; if the cold load MIC>0.6 is set, the ambient temperature, air pressure, total radiation, dry bulb temperature, wet bulb temperature, and dew point temperature are selected; similarly, if the air load MIC>0.6 is set, the ambient temperature, air pressure, dry bulb temperature, wet bulb temperature, and dew point temperature are selected.

[0093] In the multi-load cross-correlation, the correlation values ​​between electric load and heat load, cooling load and gas load are 0.5566, 0.7407 and 0.7188 respectively, the correlation values ​​between heat load and cooling load and gas load are 0.7867 and 0.7671 respectively, and the correlation value between cooling load and gas load is 0.8967. By setting the multi-load correlation threshold to 0.7, cooling load and gas load can be selected as the input of electric load and heat load prediction data set, and heat, gas and electric load can be selected as the input of cooling load and gas load prediction data set.

[0094] In the multivariate load autocorrelation results, as the prediction time moves forward (historical time), the correlation between the predicted value and the historical value shows a periodicity, that is, it decreases successively and the load value at the same time is too large. The thresholds set for the ACF of electricity, heat, cooling, and gas loads are 0.85, 0.9, 0.95, and 0.9, respectively, and the selected historical load values ​​are used as the input of the prediction data set.

[0095] A prediction data set is constructed based on the features screened out by the comprehensive analysis of the previous features and the output load, and a prediction data set without feature screening is set up.

[0096] A day-ahead forecasting comparison experiment was conducted under the same forecasting model, focusing only on the electric load. The experimental comparison results are as follows: Figures 2 to 5 shown.

[0097] The specific prediction performance indicators of other loads are shown in Tables 1 and 2.

[0098] Table 1. Day-ahead prediction accuracy - without feature selection

[0099]

[0100]

[0101]

[0102]

[0103] For electrical load, Figure 2 , 3 , 4, and 5 show that before the feature comprehensive analysis method is integrated with the hybrid deep learning model, the model prediction value does not fit the true value very well, especially the error is large at the peak. After the feature comprehensive analysis method is integrated with the hybrid deep learning model, the prediction effect is better and the error is smaller.

[0104] As can be seen from Table 1-2, the prediction experimental results before and after feature screening show that the MAE, RMSE, and MAPE before screening are larger than those after screening, and after comprehensive feature analysis, the prediction accuracy of electricity, heat, cooling, and gas loads are improved by 11.55%, 12.29%, 10.32%, and 13.28%, respectively.

[0105] Based on the above experimental results, it can be concluded that the feature comprehensive correlation analysis and hybrid deep learning prediction method described in the present invention has the best prediction performance in multivariate load prediction in IES.

[0106] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for predicting power load, characterized in that: include: Obtain multivariate load data and conduct comprehensive correlation analysis on the multivariate load data to obtain high correlation features; The comprehensive correlation analysis includes correlation analysis between meteorological factors and multiple loads, coupling analysis between multiple loads, and correlation analysis of multiple loads. A hybrid deep learning prediction model is constructed by integrating convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms; The highly correlated features are input into a hybrid deep learning prediction model for prediction to obtain predicted multivariate load values.

2. The power load forecasting method according to claim 1, characterized in that: The correlation analysis between meteorological factors and multivariate loads is performed through the maximum mutual information coefficient, which is expressed as: Among them, MIC(X,Y) represents the MIC value of features X and Y; p(x) and p(y) represent the marginal probability distribution functions of features X and Y respectively; p(x,y) represents the joint probability distribution function of features X and Y; a and b represent the number of grids divided in the x and y directions in the two-dimensional space; B is a constant.

3. The power load forecasting method according to claim 2, characterized in that: The coupling analysis between multiple loads was performed using the Spearman correlation coefficient, which is expressed as: Among them, ρ x,y represents the correlation between the load and the input feature, ρ x,y ≤1, the closer to 1, the higher the correlation; x i ,y i is the value of the two factors of the ith data point; is called the mean of the two factors; n is the number of data points.

4. The power load forecasting method according to claim 3, characterized in that: The correlation analysis of multiple loads is performed through the autocorrelation function, which is expressed as: Among them, k, ρ k represent the lag order and the ACF value of lag k respectively; t , O t-k They represent the load at time t and time tk respectively; Represents the average value of all loads.

5. The power load forecasting method according to claim 4, characterized in that: Inputting the highly correlated features into the hybrid deep learning prediction model for prediction includes: The convolutional neural network layer extracts spatial features from the input high-correlation feature data to obtain a spatial feature set; The spatial feature set is input into the BiLSTM layer. The BiLSTM layer trains multiple LSTM units from the forward and backward directions respectively. The forward and backward information features of the data before a certain moment of the spatial feature set are read simultaneously through the hidden layer to obtain the relationship between the current moment data and the past and future moment data. The current moment data is fitted from two directions to obtain the power load data at a certain moment in the future.

6. The power load forecasting method according to claim 5, characterized in that: Inputting the highly correlated features into the hybrid deep learning prediction model for prediction, further comprising: The attention mechanism layer assigns different weights to each feature, and the weights are used to calculate the attention score through the softmax function; The weighted features are input into the fully connected layer for prediction, and the predicted multivariate load value is output.

7. The power load forecasting method according to claim 6, characterized in that: The attention score is obtained by calculating the similarity between the query vector and each input vector. There are n attention scores represented as: s1, s2, s3, ..., s n ; The Softmax function converts the original attention score into a weight in the form of a probability distribution. For any attention score s i , and its weight calculation formula after softmax is: in, Represents the attention score s i Take the index, is the sum of all attention scores after exponentiation; The calculated weights are used to perform weighted summation on the corresponding input elements. The output vector after attention weighting is expressed as {v1,v2,…,v n }, the output is: Among them, ω i is the corresponding weight.

8. A system using the power load forecasting method according to claim 1, characterized in that: include: An acquisition module is used to acquire multivariate load data and conduct a comprehensive correlation analysis on the multivariate load data to obtain high correlation features; The comprehensive correlation analysis includes the correlation analysis between meteorological factors and multiple loads, the coupling analysis between multiple loads and the correlation analysis of multiple loads. Building blocks for constructing hybrid deep learning prediction models by integrating convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms; The prediction module is used to input the high correlation features into the hybrid deep learning prediction model for prediction to obtain the predicted multivariate load value.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the power load forecasting method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the power load forecasting method according to any one of claims 1 to 7.

Images