A power load prediction method and device for high proportion of new energy access

Abstract

The application discloses a power load prediction method and device for high-proportion new energy access, which comprises the following steps: adopting a Pearson correlation coefficient and a principal component analysis method to perform correlation analysis and nonlinear dimension reduction processing on multi-dimensional features of labeled power load data, and screening key features; defining a new energy output fluctuation factor to quantize the lag error of new energy output fluctuation load; using wavelet decomposition to decompose total load into basic load and new energy output fluctuation load, and calculating short-term fluctuation factors and long-term fluctuation factors; constructing training data sets by using the key features of the labeled power load data and corresponding short-term fluctuation factors and long-term fluctuation factors; training a pre-constructed power load prediction model based on CNN-LSTM by using the training data sets, and obtaining a trained power load prediction model; and predicting power load for high-proportion new energy access by using the trained power load prediction model.

Description

Technical Field

[0001] This application relates to a power load forecasting method and apparatus for high-proportion renewable energy access, belonging to the field of power load forecasting technology. Background Technology

[0002] With the rapid development of smart grids and the energy internet, power load forecasting has become a core technology for power system dispatch optimization, energy supply and demand balance, and grid security. Accurate load forecasting can effectively reduce generation costs, improve resource utilization, and provide key decision support for electricity market transactions.

[0003] Current power load forecasting technology faces several challenges: With the large-scale integration of renewable energy into the grid, the volatility of renewable energy output leads to strong randomness in net load (the difference between total load and renewable energy output). Traditional power load forecasting methods rely solely on historical load and meteorological data, failing to quantify the dynamic impact of renewable energy output forecasting errors. This results in accumulated forecasting biases, making it difficult to respond promptly to sudden changes in renewable energy output and potentially causing grid frequency instability or wind and solar power curtailment. Furthermore, raw power load data is multi-source and heterogeneous, encompassing meteorological and user behavior characteristics. Traditional feature selection methods easily overlook dynamic correlations, affecting model input. Actual load data often exhibits abnormal fluctuations (e.g., equipment failure, human error) and random missing data (e.g., sensor communication interruptions). High-dimensional features are strongly coupled, leading to significant information loss in traditional linear dimensionality reduction and a tendency to overfit using raw data.

[0004] Existing prediction models mostly rely on a single deep learning architecture (such as pure LSTM or CNN). While CNN can extract local spatial features, it struggles to capture long-term dependencies and is not dynamically coupled with the prediction error of new energy sources. LSTM excels at time series modeling but is insensitive to spatial correlations and fails to quantify the impact of fluctuations in new energy output. Furthermore, unoptimized data preprocessing and feature engineering further limit the model's ability to learn dynamic load characteristics.

[0005] Therefore, there is an urgent need for a load forecasting method that integrates efficient data processing, dynamic feature optimization, and multimodal deep learning to address the load forecasting deviation caused by power output fluctuations in high-proportion renewable energy power grids. Summary of the Invention

[0006] Objective: In view of at least one of the above technical problems, this application provides a power load forecasting method and device for high-proportion renewable energy access, which solves the problem of load forecasting deviation caused by power output fluctuation in high-proportion renewable energy power grids by introducing a renewable energy output fluctuation factor.

[0007] The technical solution adopted in this application is:

[0008] Firstly, this application provides a power load forecasting method for high-proportion renewable energy integration, including:

[0009] S1. Pearson correlation coefficient was used to perform correlation analysis on the multidimensional features of labeled power load data to screen preliminary features;

[0010] S2. Principal component analysis is used to perform nonlinear dimensionality reduction on the preliminary features of the power load data to screen out the key features of the power load data; the power load data includes the total load at each time point and the corresponding temperature, humidity and electricity price;

[0011] S3. Define the new energy output fluctuation factor and quantify the lag error of the new energy output fluctuation load; use wavelet decomposition to decompose the total load into basic load and new energy output fluctuation load, and calculate the short-term fluctuation factor and long-term fluctuation factor in combination with the definition of the new energy output fluctuation factor.

[0012] S4. Construct a training dataset from the key features of labeled power load data and the corresponding short-term and long-term fluctuation factors. Use the training dataset to train the pre-built CNN-LSTM-based power load prediction model to obtain the trained power load prediction model. The power load prediction model includes a CNN module, an LSTM module, and a fully connected layer. Concatenate the key features of the power load data with the short-term fluctuation factors to form a continuous feature map, which is used as the input to the CNN module. The CNN module extracts the feature sequence. Concatenate the feature sequence with the long-term fluctuation factors and input it into the LSTM module in the form of a time series for prediction, outputting a prediction vector. The prediction vector is then passed through a fully connected layer to output the power load prediction result.

[0013] S5. Using the trained power load prediction model, the power load for high-proportion renewable energy access is predicted to obtain the power load prediction results.

[0014] Secondly, this application provides a power load forecasting device for high-proportion renewable energy access, including a processor and a storage medium;

[0015] The storage medium is used to store instructions;

[0016] The processor is configured to operate according to the instructions to execute the method according to the first aspect.

[0017] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0018] Fourthly, this application provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect.

[0019] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect.

[0020] Beneficial Effects: The power load forecasting method and apparatus for high-proportion renewable energy access provided in this application have the following advantages: Addressing the multi-source heterogeneous nature of power load data, this application introduces Pearson correlation analysis and principal component analysis. By establishing a time-series correlation matrix, key features are accurately screened, renewable energy output fluctuation factors are defined, and the lag error of renewable energy output fluctuation load is quantified. A normalized dynamic embedding prediction model is achieved by separating the base load through wavelet decomposition. A hybrid model of convolutional neural network (CNN) and long short-term memory network (LSTM) is constructed. The obtained training dataset is concatenated with short-term fluctuation factors to construct a continuous feature map, and then the CNN is used to extract feature sequences. The feature sequences are concatenated with long-term fluctuation factors and load fluctuation components and input into the LSTM network in time series form for prediction, outputting a prediction vector. The CNN is responsible for extracting short-term fluctuations and key local spatial features, while the LSTM models long-term temporal dependencies, achieving more accurate power load forecasting for high-proportion renewable energy access. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating a power load forecasting method for high-proportion renewable energy access according to an embodiment of this application.

[0022] Figure 2 This is a schematic diagram of the training process for an electricity load prediction model according to one embodiment of this application.

[0023] Figure 3 This is a schematic diagram of a CNN convolutional neural network structure according to an embodiment of this application.

[0024] Figure 4 This is a schematic diagram of an LSTM (Long Short-Term Memory) neural network structure according to an embodiment of this application. Detailed Implementation

[0025] The present application will be further described below with reference to the accompanying drawings and embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present application, and should not be used to limit the scope of protection of the present application.

[0026] In the description of this application, "several" means one or more, "multiple" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0027] In the description of this application, the terms "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0028] The term "and / or" simply describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0029] Example 1: This example provides a power load forecasting method for high-proportion renewable energy integration, such as... Figure 1 As shown, it includes:

[0030] S1. Preprocess the raw power load data and use the Pearson correlation coefficient to perform correlation analysis on the multidimensional features of the labeled power load data to screen preliminary features.

[0031] Furthermore, before performing correlation analysis on the multidimensional characteristics of the power load data using the Pearson correlation coefficient, the process also includes: preprocessing the power load data, specifically including:

[0032] The power load data is processed by column to handle missing and outlier values, and then the data is normalized. The missing value processing operation uses linear interpolation to fill missing data; the outlier processing method is to replace outlier data with vertical correction; the normalization operation uses the min-max method to normalize the feature data.

[0033] In step S1, the Pearson correlation coefficient is used to perform correlation analysis on the multidimensional characteristics of the power load data to screen preliminary features, including:

[0034] Let the input feature variables of the power load forecasting model be... The output is the power load forecast result. The Pearson correlation coefficient is expressed as:

[0035]

[0036] In the formula: This represents the Pearson correlation coefficient; Indicates the number of feature variables; Pearson correlation coefficient measures... , Linear relationship, with values ​​in the range of [-1, 1]. When the absolute value of the Pearson correlation coefficient approaches 1, it indicates a strong linear correlation between the feature and the power load forecast results; when the Pearson correlation coefficient approaches 0, it reflects a weak statistical dependence between the feature and the power load forecast results.

[0037] Calculate the Pearson correlation coefficient between the multidimensional features of power load data and the power load forecast results; select features with a Pearson correlation coefficient greater than a set threshold as preliminary features obtained through screening.

[0038] S2. Principal component analysis (PCA) is used to perform nonlinear dimensionality reduction on the preliminary features of the power load data to screen out the key features of the power load data; the power load data includes the total load at each time point as well as the corresponding temperature, humidity and electricity price.

[0039] In step S2, principal component analysis is used to perform nonlinear dimensionality reduction on the preliminary features of the power load data, and key features of the power load data are selected, including:

[0040] Arrange the preliminary characteristics of all power load data into a matrix and zero-mean by row, and set... Dimensional Samples Each sample has Features Calculate the covariance matrix And calculate the covariance matrix. The various eigenvalues and the corresponding feature vectors ; , Indicates the first 1 eigenvalue, Indicates the first 1 eigenvector This represents the number of eigenvalues ​​in the covariance matrix;

[0041] ,

[0042] In the formula, Represents covariance operation;

[0043] Calculate the variance contribution rate of each eigenvalue of the covariance matrix;

[0044] ,in Indicates the first Variance contribution rate of each eigenvalue;

[0045] Arrange the eigenvectors into a matrix according to the variance contribution rate of their corresponding eigenvalues ​​in descending order, and select the matrix that satisfies the cumulative contribution rate. Size setting value The first of them The eigenvector matrix is ​​composed of eigenvectors. , to achieve dimensionality reduction Data after dimension , This represents the power load data after initial feature filtering;

[0046] Cumulative contribution rate The calculation method is as follows:

[0047] .

[0048] In this embodiment, the key features for obtaining the power load data are temperature, humidity, and electricity price.

[0049] S3. Define the new energy output fluctuation factor and quantify the lag error of the new energy output fluctuation load; use wavelet decomposition to decompose the total load into basic load and new energy output fluctuation load, and calculate the short-term fluctuation factor and long-term fluctuation factor in combination with the definition of the new energy output fluctuation factor.

[0050] In some embodiments, step S3 specifically includes:

[0051] The total load is decomposed into basic loads using wavelet decomposition. And fluctuations in power output from new energy sources ;

[0052] ,

[0053] In the formula, For total load, Based on the load, Fluctuating loads for renewable energy output;

[0054] Define a new energy power output fluctuation factor to quantify the lag error of new energy power output fluctuation load;

[0055] ,

[0056] In the formula, Fluctuation factors in the contribution of new energy sources To predict the amount of time lag compensation, for The real value of the fluctuation in renewable energy output load at any given time. for The real value of the fluctuating load of new energy power output at any given time;

[0057] Based on the time scale, the minute-level fluctuation factor of new energy output is defined as the short-term fluctuation factor. The average of the short-term volatility factor calculated using a 24-hour moving average is defined as the long-term volatility factor. It is obtained by extracting daily trends;

[0058] ,

[0059] In the formula, for The fluctuation factor of new energy output at any given time; Indicates to The fluctuation factors of new energy output at each time point within the 24 hours prior to the current time are summed.

[0060] S4. Construct a training dataset from the key features of the labeled power load data and the corresponding short-term and long-term fluctuation factors. Use the training dataset to train the pre-built CNN-LSTM-based power load prediction model to obtain the trained power load prediction model.

[0061] It should be noted that, as Figure 2 As shown, the power load prediction model includes a CNN module, an LSTM module, and a fully connected layer. The key features of the power load data are concatenated with short-term fluctuation factors to form a continuous feature map, which is used as the input to the CNN module. The CNN module extracts the feature sequence. The feature sequence is concatenated with long-term fluctuation factors and input into the LSTM module in the form of a time series for prediction, and the prediction vector is output. The prediction vector is then passed through the fully connected layer to output the power load prediction result.

[0062] Furthermore, such as Figure 3 As shown, the CNN module adopts a three-tiered convolutional structure, with each layer configured as a composite unit consisting of alternating stacked Conv1d and MaxPooling1D layers. Specific parameters are designed as follows: the convolutional kernel size is set to 3, the stride is maintained at 1, and each convolutional unit is configured with two independent convolutional kernels to extract multi-scale temporal features. The pooling window size is consistent with the convolutional kernel size, and the data dimensionality is gradually reduced through three downsampling operations. During feature extraction, each convolutional layer is followed by a pooling layer with the same parameter configuration, forming an alternating "convolution-pooling" feature compression mechanism. Finally, a global average pooling layer flattens the three-dimensional feature map into a one-dimensional feature vector.

[0063] Furthermore, such as Figure 4 As shown, the LSTM module is set to 4 layers, with 64 neurons in each layer, and the output of the LSTM is a prediction vector.

[0064] In some embodiments, different weights are assigned to the time-series characteristics and fluctuation factors of the power load data, and the weighted mean square error (WMSE) is introduced as a loss function during the training of the power load forecasting model to improve the prediction accuracy of the model during critical periods. The calculation process of WMSE is as follows;

[0065]

[0066] In the formula: and The first The true and predicted values ​​of each sample The number of samples.

[0067] S5. Using the trained power load prediction model, the power load for high-proportion renewable energy access is predicted to obtain the power load prediction results.

[0068] More specifically, for power load data oriented towards high proportion of renewable energy access, key features are first extracted; then wavelet decomposition is used to decompose the total load into basic load and renewable energy output fluctuation load, and short-term fluctuation factor and long-term fluctuation factor are calculated; then the key features of the power load data and the corresponding short-term fluctuation factor and long-term fluctuation factor are input into the trained power load prediction model to predict the power load prediction results.

[0069] In some embodiments, the data processing procedure for the CNN-LSTM-based power load forecasting model includes:

[0070] Key characteristics of power load data and short-term fluctuation factors The concatenated feature maps are used as input to a CNN module, which then extracts the feature sequence. ;

[0071] ;

[0072] In the formula For short-term fluctuation factors, T, H, and E are the key features obtained through screening, which are temperature, humidity, and electricity price, respectively;

[0073] feature sequence With long-term volatility factors The concatenation and time series data are input into the LSTM module for prediction, and the output is a prediction vector.

[0074] The prediction vector is passed through a fully connected layer to output the power load prediction result.

[0075] In this embodiment, the power load forecasting results are output in a specified form, and the method of this application enables more accurate power load forecasting for high proportions of renewable energy access.

[0076] Example 2: Based on Example 1, this example provides a power load forecasting device for high-proportion renewable energy access, including a processor and a storage medium;

[0077] The storage medium is used to store instructions;

[0078] The processor is configured to operate according to the instructions to execute the method according to Embodiment 1.

[0079] Example 3: Based on Example 1, this example provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in Example 1.

[0080] Example 4: Based on Example 1, this example provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described in Example 1.

[0081] Example 5: Based on Example 1, this example provides a computer program product, including a computer program that, when executed by a processor, implements the method described in Example 1.

[0082] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0083] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0084] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0085] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0086] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A power load forecasting method for high-proportion renewable energy integration, characterized in that, include: S1. Pearson correlation coefficient was used to perform correlation analysis on the multidimensional features of labeled power load data to screen preliminary features; S2. Principal component analysis is used to perform nonlinear dimensionality reduction on the preliminary features of the power load data to screen out the key features of the power load data; the power load data includes the total load at each time point and the corresponding temperature, humidity and electricity price; S3. Define the new energy output fluctuation factor and quantify the lag error of the new energy output fluctuation load; Wavelet decomposition is used to decompose the total load into basic load and renewable energy output fluctuation load. Short-term and long-term fluctuation factors are calculated based on the definition of renewable energy output fluctuation factor, including: The total load is decomposed into basic loads using wavelet decomposition. Fluctuating load of new energy output ; In the formula, For total load, Based on the load, Fluctuating loads for renewable energy output; Define a new energy power output fluctuation factor to quantify the lag error of new energy power output fluctuation load; In the formula, Fluctuation factors in the contribution of new energy sources To predict the amount of time lag compensation, for The real value of the fluctuation in renewable energy output load at any given time. for The real value of the fluctuating load of new energy power output at any given time; Based on the time scale, the minute-level fluctuation factor of new energy output is defined as the short-term fluctuation factor. The average of the short-term volatility factor calculated using a 24-hour moving average is defined as the long-term volatility factor. It is obtained by extracting daily trends; In the formula, for The fluctuation factor of new energy output at any given time; Indicates to The fluctuation factors of new energy output at various times within the 24 hours prior to the current time are summed. S4. Construct a training dataset from the key features of labeled power load data and the corresponding short-term and long-term fluctuation factors. Use the training dataset to train the pre-built CNN-LSTM-based power load prediction model to obtain the trained power load prediction model. The power load prediction model includes a CNN module, an LSTM module, and a fully connected layer. Concatenate the key features of the power load data with the short-term fluctuation factors to form a continuous feature map, which is used as the input to the CNN module. The CNN module extracts the feature sequence. Concatenate the feature sequence with the long-term fluctuation factors and input it into the LSTM module in the form of a time series for prediction, outputting a prediction vector. The prediction vector is then passed through a fully connected layer to output the power load prediction result. S5. Using the trained power load prediction model, the power load for high-proportion renewable energy access is predicted to obtain the power load prediction results.

2. The method according to claim 1, characterized in that, S1. Using Pearson correlation coefficient to perform correlation analysis on the multidimensional characteristics of power load data to screen preliminary features, including: Let the input feature variables of the power load forecasting model be... The output is the power load forecast result. The Pearson correlation coefficient is expressed as: ； In the formula: This represents the Pearson correlation coefficient; Indicates the number of feature variables; Pearson correlation coefficient measures... , Linear relationship, with values ​​in the range of [-1, 1]. When the absolute value of the Pearson correlation coefficient approaches 1, it indicates a strong linear correlation between the feature and the power load forecast results; when the Pearson correlation coefficient approaches 0, it reflects a weak statistical dependence between the feature and the power load forecast results. Calculate the Pearson correlation coefficient between the multidimensional features of power load data and the power load forecast results; select features with a Pearson correlation coefficient greater than a set threshold as preliminary features obtained through screening.

3. The method according to claim 1, characterized in that, S2. Principal component analysis is used to perform nonlinear dimensionality reduction on the preliminary features of the power load data, and key features of the power load data are selected, including: Arrange the preliminary characteristics of all power load data into a matrix and zero-mean by row, and set... Dimensional Samples Each sample has Features Calculate the covariance matrix And calculate the covariance matrix. The various eigenvalues and the corresponding feature vectors ; , Indicates the first 1 eigenvalue, Indicates the first 1 eigenvector This indicates the number of eigenvalues ​​in the covariance matrix; ， In the formula, Represents covariance operation; Calculate the variance contribution rate of each eigenvalue of the covariance matrix; ,in Indicates the first Variance contribution rate of each eigenvalue; Arrange the eigenvectors into a matrix according to the variance contribution rate of their corresponding eigenvalues ​​in descending order, and select the matrix that satisfies the cumulative contribution rate. Size setting value The first of them The eigenvector matrix is ​​composed of eigenvectors. , to achieve dimensionality reduction Data after dimension , This represents the power load data after initial feature filtering; Cumulative contribution rate The calculation method is as follows: 。 4. The method according to claim 1, characterized in that, The CNN module adopts a three-level convolutional structure, with each layer configured with a composite unit of Conv1d layer and MaxPooling1D layer stacked alternately; the specific parameters are designed as follows: the convolutional kernel size is set to 3, the stride is kept to 1, and each level of convolutional unit is configured with 2 independent convolutional kernels to extract multi-scale temporal features. The pooling window size is consistent with the convolution kernel, and the data dimensionality is gradually reduced through three downsampling operations; During the feature extraction process, each convolutional layer is followed by a pooling layer with the same parameter configuration, forming an alternating feature compression mechanism of "convolution-pooling". Finally, the three-dimensional feature map is flattened into a one-dimensional feature vector through a global average pooling layer.

5. The method according to claim 1, characterized in that, The LSTM module is set to 4 layers, with 64 neurons in each layer, and the output of the LSTM is a prediction vector.

6. The method according to claim 1, characterized in that, The data processing steps for the CNN-LSTM-based power load forecasting model include: Key characteristics of power load data and short-term fluctuation factors The concatenated feature maps are used as input to a CNN module, which then extracts the feature sequence. ; ； In the formula For short-term fluctuation factors, T, H, and E are the key features obtained through screening, which are temperature, humidity, and electricity price, respectively; feature sequence With long-term volatility factors The concatenation and time series data are input into the LSTM module for prediction, and the output is a prediction vector. The prediction vector is passed through a fully connected layer to output the power load prediction result.

7. The method according to claim 1, characterized in that, Before using the Pearson correlation coefficient to perform correlation analysis on the multidimensional characteristics of the power load data, the following steps are also included: preprocessing the power load data, specifically including: The power load data is processed by column to handle missing and outlier values, and then the data is normalized. The missing value processing operation uses linear interpolation to fill missing data; the outlier processing method is to replace outlier data with vertical correction; the normalization operation uses the min-max method to normalize the feature data.

8. A power load forecasting device for high-proportion renewable energy integration, characterized in that, Including processor and storage media; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method according to any one of claims 1 to 7.

Images