GWO-VMD-NLM time sequence analysis method for improving power load prediction accuracy

Abstract

According to the GWO-VMD-NLM time sequence analysis method for improving the power load prediction accuracy, key parameters in the variational mode decomposition process are automatically selected by improving a grey wolf optimization algorithm, and then a power load sequence is decomposed into a series of subsequences through VMD after parameter optimization; the method comprises the following steps of: acquiring a plurality of sub-sequences, introducing internal regularity and complexity of sample entropy capture sub-sequences, recombining the sub-sequences into a low-complexity group F1 and a high-complexity group F2, and finally performing non-local mean filtering on the high-complexity group F2 to further filter noise and disturbance signals so as to improve the quality of power load data input into a prediction model. And the accuracy of subsequent power load prediction is improved. According to the method, different frequency components and trends in the power load time sequence are effectively captured, the load sequence with rich features is extracted, the recombined load sequence can be input into various power load prediction models, and the accuracy of subsequent power load prediction is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system data analysis, and specifically to a GWO-VMD-NLM time series analysis method for improving the accuracy of power load forecasting. Background Technique

[0002] Building a new power system with new energy as the main body is an important prerequisite and inevitable trend for promoting the low-carbon transformation and development of modern power systems. As an important technical means for guiding the dispatching of the new power system, power load forecasting can not only reduce energy waste and grid failures by accurately predicting power demand, but also improve the stability of the power grid. The analysis of power load time series is a key link in power load forecasting. Whether the power load time series data can be effectively analyzed will significantly affect the forecasting effect.

[0003] Many traditional power load time series analysis methods usually require the data to be stationary, while power load data is usually non-stationary, that is, its statistical characteristics such as mean and variance change with time, and the load changes of the power system are often affected by various non-linear factors such as weather conditions, holiday effects, and emergencies. Therefore, complex preprocessing steps such as differencing and transformation are required, but these steps may lead to information loss, resulting in poor performance of the forecasting model in power load forecasting. At the same time, with the transformation of traditional power grids to smart grids, power load data is continuously accumulating, and the data acquisition level of external influences such as meteorology is continuously improving. Processing these high-dimensional data requires more complex data processing technologies, while traditional methods are often difficult to effectively extend to high-dimensional data analysis, and require carefully selected parameter settings to achieve the best performance. The selection of these parameters often depends on expert experience or repeated experiments, which is inefficient and difficult to optimize. Generally speaking, the stable operation of the new power system depends on the accurate forecasting and analysis of power loads. Traditional power load analysis methods usually face the challenges of dealing with non-linear and non-stationary data, which limits the accuracy and reliability of forecasting.

[0004] Technical comparison with the patent CN113935513A "A short-term power load forecasting method based on CEEMDAN"

[0005] In Patent CN113935513A, CEEMDAN is used to decompose the original electrical load sequence into several IMF components and a trend component. However, in our method, variational mode decomposition (VMD) based on improved grey wolf optimization (GWO) is used to decompose the original electrical load sequence. The Akaike information criterion (AIC) is utilized to establish the objective function of GWO in combination with the residual sum of squares (RSS). By minimizing the objective function of GWO, the key parameters α and K in the VMD process are automatically optimized and updated iteratively until the optimal α and K are obtained. These are determined as the final VMD parameters, and the original electrical load sequence is decomposed by VMD to obtain a series of power load subsequences (IMFs).

[0006] In Patent CN113935513A, the sample entropy is used to measure the noise level of each component, i.e., the IMF and the trend component. Several IMF components and trend components with a sample entropy difference within 0.2 are superimposed to form a new subsequence. In this study, the sample entropy values of each IMF are calculated, and the quartiles of the sample entropy values of each IMF are selected as the thresholds. Each IMF is recombined into a low-complexity group and a high-complexity group, and the high-complexity group is further filtered by the non-local means algorithm to remove the noise and disturbances in the high-complexity group.

[0007] Technical comparison with Patent CN118157129A, "A Power Plant Generation Prediction Method Based on a Hybrid Prediction Framework"

[0008] In Patent CN118157129A, each perturbation signal is decomposed by CEEMD to obtain N IMF components. However, in our method, VMD based on improved GWO is used to decompose the original electrical load sequence. The Akaike information criterion (AIC) is utilized to establish the objective function of GWO in combination with the residual sum of squares (RSS). By minimizing the objective function of GWO, the key parameters α and K in the VMD process are automatically optimized and updated iteratively until the optimal α and K are obtained. These are determined as the final VMD parameters, and the original electrical load sequence is decomposed by VMD to obtain a series of power load subsequences (IMFs).

[0009] In Patent CN118157129A, based on similarity features, the sample entropy of N IMF sequence components is calculated. The N original IMF sequence components are reconstructed into M IMF sequence components, and the sum of the M sequence components is obtained to get the final IIMF sequence. In this study, the sample entropy values of each IMF obtained by VMD are calculated, and the quartiles of the sample entropy values of each IMF are selected as the thresholds. Each IMF is recombined into a low-complexity group and a high-complexity group, and the high-complexity group is further filtered by the non-local means algorithm to remove the noise and disturbances in the high-complexity group.

[0010] In view of the above problems, the present invention aims to propose a GWO-VMD-NLM time series analysis method for improving the accuracy of electric load forecasting. Summary of the Invention

[0011] To solve the above technical problems, the present invention proposes a GWO-VMD-NLM time series analysis method for improving the accuracy of electric load forecasting. This method aims to use an improved grey wolf algorithm to optimize the key parameters in the VMD process to decompose the electric load sequence, and introduce sample entropy to recombine the decomposed electric load sequences into a low-complexity group F1 and a high-complexity group F2. Perform non-local means filtering (NLM) on the high-complexity group F2 to further filter out noise and disturbance signals, so as to improve the quality of the electric load data input into the prediction model, and further improve the accuracy of subsequent electric load forecasting.

[0012] To achieve the above object, the technical solution adopted by the present invention is:

[0013] The GWO-VMD-NLM time series analysis method for improving the accuracy of electric load forecasting is specifically as follows:

[0014] Step 101: Select the electric load time series data to be processed and perform necessary preprocessing on the data.

[0015] Step 102: Introduce the Akaike information criterion into the grey wolf optimization algorithm to automatically optimize and determine the key parameters of VMD: the secondary penalty factor α and the number of modes K. Use the optimized VMD algorithm to decompose the selected electric load time series data into multiple subsequences.

[0016] Step 103: Calculate the correlation coefficients between the decomposed subsequences and between each decomposed subsequence and the original load sequence to evaluate the effectiveness of the decomposition.

[0017] Step 104: Introduce sample entropy, and recombine the load sequence into a low-complexity group F1 and a high-complexity group F2 according to the sample entropy of each subsequence. Evaluate the effectiveness of the reconstruction through the correlation coefficients between the recombined sequences and between each recombined sequence and the original load sequence.

[0018] Step 105: Perform non-local means filtering on the high-complexity group to further remove the noise of the electric load sequence while retaining important features.

[0019] Step 106: Input the electric load time series before and after being processed by this method into the same model for load forecasting.

[0020] Furthermore, the specific steps of Step 101 are as follows;

[0021] Process the missing values in the power load time series. Considering the non-linearity of the power load time series, the second-order polynomial interpolation method is used to fill the missing values. Specifically, if the value of x i is missing or abnormal, then the first two sample points x i-1 , x i-2 and the last two sample points x i+1 and x i+2 of the missing value are substituted into formula (1) to solve for the missing value of x i ;

[0022]

[0023] Furthermore, the specific steps of step 102 are as follows;

[0024] In step 102, the parameters of GWO and VMD are first initialized. The preprocessed power load sequence is decomposed by VMD using the initialized parameters α and K to obtain a set of power load subsequences. Then, the Akaike information criterion is used to establish the objective function of GWO, and the expression is calculated as follows:

[0025]

[0026] Among them, <α, K> are the secondary penalty factor and the number of modal decompositions of the VMD to be optimized, n is the number of sample points of the load sequence, and RSS is shown in formula (3):

[0027]

[0028] Among them, y i and are the i-th values of the original load sequence and the reconstructed load sequence respectively, is calculated as shown in formula (4)

[0029]

[0030] Among them, u k (t) is the k-th modal component obtained by VMD decomposition;

[0031] The α and K are automatically optimized and updated through the objective function of GWO and iterated continuously until the optimal α and K are obtained. They are determined as the final VMD parameters and the L is decomposed by VMD to obtain a series of power load subsequences.

[0032] Furthermore, the specific steps of step 103 are as follows;

[0033] Calculate the correlation coefficient matrix between the IMFs of each subsequence. If the absolute values of all correlation coefficients are less than 0.25, it proves that the decomposition effect is good;

[0034] Further, the specific steps of step 104 are as follows;

[0035] First, calculate the sample entropy of each IMF. The sample entropy is defined as:

[0036]

[0037] where B m (r) is the probability that two sequences match at m points under the similarity tolerance r, and A m (r) is the probability that two sequences match at m + 1 points. Then, based on the sample entropy results, the IMFs are recombined into two sequences, specifically:

[0038] Calculate the sample entropy of all IMFs. Use the quartile of the sample entropy as the threshold for distinguishing the recombined sequences. Classify the subsequences with sample entropy lower than the threshold into the low-complexity group F1, which usually contains more regular and periodic signal components. Classify the subsequences with sample entropy higher than the threshold into the high-complexity group F2, which usually contains more complex or noisy components. Evaluate the effectiveness of the recombination by the correlation coefficients between the recombined sequences F1, F2, and L.

[0039] Further, the specific steps of step 105 are as follows;

[0040] Further denoise the recombined high-complexity group F2 by the non-local means filtering algorithm. The principle of NLM is as follows:

[0041]

[0042] where u(t) is the noisy signal, v(t) is the original signal, and n(t) is the noise interference signal. The average estimated value of sample x is obtained by calculating the weighted set of different points t in the adjacent region:

[0043]

[0044]

[0045] where w(x, t) is the weight, v(x) is the output at sample point x, and z(x) is the accumulation of weights in the search domain. The weight calculation is shown in formulas (9) and (10):

[0046]

[0047]

[0048] where x, t are sample points, d represents the Gaussian weighted Euclidean distance, N(x) is the neighborhood block centered on x, and λ is the parameter affecting the smoothness of the filtered signal.

[0049] Further, the specific steps of step 106 are as follows;

[0050] Select a prediction model to predict the original power load time series and the power load time series processed by this method in the same environment, and use the four indicators of 2 goodness of fit, MAE (mean absolute error), RMSE (root mean square error), and MAPE (mean absolute percentage error) for evaluation to verify the effectiveness of this method for subsequent power load series prediction.

[0051] The present invention adopts the above technical solutions and has the following beneficial effects:

[0052] (1) The present invention automatically adjusts the key parameters α and the number of modes K of VMD through the grey wolf optimization algorithm. This parameter optimization process is automatic, reducing the need for manual intervention and subjective judgment, improving the objectivity and accuracy of the analysis, and thus obtaining a more accurate decomposition result.

[0053] (2) The present invention combines the number of modes K and the residual RSS as the objective function of the grey wolf optimization algorithm using the Akaike information criterion (AIC). During the optimization process, both the goodness of fit and complexity after VMD decomposition are considered, ensuring that the decomposition result can not only fit the data well but also not be overly complex, thereby improving the generalization ability of the decomposition. In addition, by calculating the correlation coefficient matrix of each IMF component after decomposition, the decomposition effect can be evaluated more objectively.

[0054] (3) The present invention proposes a method for recombining power load time series based on sample entropy, which recombines the sequence into a low-complexity group and a high-complexity group, helping to extract the most informative part from complex signals. At the same time, non-local means filtering (NLM) is proposed for the high-complexity group to further remove noise in the power load sequence while retaining important features.

[0055] (4) The power load series generated by the present invention can be input into various power load prediction models. Through the decomposition and recombination of the power load data series, it can effectively capture different frequency components and trends in the data, and extract a load sequence with rich features, having a good effect on subsequent power load prediction. Description of the Drawings

[0056] Figure 1 It is the overall flowchart of the power load time series analysis method adopted by the present invention;

[0057] Figure 2 It is the specific algorithm diagram of the power load sequence decomposition proposed by the present invention;

[0058] Figure 3 It is the power load sequence decomposition result of the embodiment of the present invention;

[0059] Figure 4 The sample entropy line graph of each IMF in the embodiment of the present invention;

[0060] Figure 5 The recombinant sequence diagram in the embodiment of the present invention;

[0061] Figure 6 The graph of using the method proposed by the present invention for prediction; Detailed implementation manners

[0062] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners:

[0063] The present invention adopts a power load time series analysis method combining the Grey Wolf Optimization algorithm (GWO), Variational Mode Decomposition (VMD) and Non-Local Means algorithm (NLM) to achieve the above invention purpose. The scheme includes: data cleaning of the power load time series, VMD decomposition of the power load sequence based on the improved GWO algorithm, recombination of the load sequence through sample entropy, NLM filtering of the high complexity group, and subsequent input into the power load prediction model for effect evaluation. Through this series of steps, different frequency components and trends in the power load time series are effectively captured, and a load sequence with rich features is extracted, which is beneficial to improving the accuracy of subsequent power load prediction.

[0064] The logical implementation of the power load time series analysis method based on GWO-VMD-NLM proposed by the present invention, as Figure 1 shown, includes steps 101 to 106.

[0065] Step 101: Select the power load time series data to be processed and perform necessary preprocessing on the data

[0066] In this embodiment, the power load data set of a certain region in 2010 is selected, with a sampling point every half hour, and there are 17,521 sample points in total. The data set format is shown in Table 1. In this embodiment, the processing object is the last column (power load column). First, the missing values in the power load time series (the last column in the data set) are processed. Considering the non-linearity of the power load time series, the second-order polynomial interpolation method is used to fill the missing values. Specifically, if the value of x i is missing or abnormal, the missing value of x i-1 , x i-2 and the last two sample points x i+1 and x i+2 are used to substitute into formula (1) to solve the missing value of x i . After preprocessing the original power load time series, it is denoted as L.

[0067]

[0068] Table 1

[0069] Step 102: Optimize the parameters of VMD through an improved Grey Wolf Optimization algorithm (GWO) to obtain the decomposed power load subsequences

[0070] The specific algorithm diagram of this process is shown in Figure 2 As shown, α is the secondary penalty factor of VMD, and K is the number of modes to be decomposed. First, initialize the parameters of GWO and VMD, and use the parameters α and K obtained by random initialization to perform VMD decomposition on the power load sequence to be processed (denoted as L) to obtain a set of power load subsequences (IMFs); then use the Akaike Information Criterion (AIC) to establish the objective function of GWO, and the expression is calculated as shown in formulas (2), (3), and (4), where <α, K> are the secondary penalty factor and the number of mode decompositions of VMD to be optimized, and n is the number of sample points of the load sequence. In this embodiment, n is 17521, y i is the i-th value of L, is the i-th value of the load sequence obtained by superimposing all IMFs at each iteration, u k (t) is the k-th mode component obtained by VMD decomposition. Automatically optimize and update α and K through the objective function of GWO and continuously iterate until the best α and K are obtained and VMD decomposition is performed on L to obtain a series of power load subsequences (IMFs). In the embodiment, the obtained α is 281.5 and K is 8, Figure 3 The images of L and the subsequences (IMF1 - IMF8) obtained by this decomposition method are shown as follows.

[0071] Step 103: Calculate the correlation coefficients between the decomposed subsequences (IMFs) and between the original load sequence

[0072] Calculate the correlation coefficient matrix between each subsequence IMF. If the absolute values of all correlation coefficients are less than 0.25, it proves that the decomposition effect is good. For better display effect, the correlation coefficient matrix is drawn into a table as shown in Table 2, which shows the correlation coefficients between each IMF and between each IMF and L. It can be seen that the correlation coefficients between the IMFs obtained by decomposition in this embodiment are small and the decomposition effect is good.

[0073] IMF1 IMF2 IMF3 IMF4 IMF5 IMF6 IMF7 IMF8 L IMF1 1.0000 0.0220 0.0070 0.0065 0.0056 0.0054 0.0038 0.0023 0.5493 IMF2 0.0225 1.0000 0.0062 0.0035 0.0030 0.0029 0.0016 0.0009 0.7302 IMF3 0.0071 0.0062 1.0000 0.0083 0.0055 0.0053 0.0018 0.0010 0.3828 IMF4 0.0065 0.0035 0.0083 1.0000 0.0294 0.0185 0.0051 0.0022 0.1758 IMF5 0.0056 0.0030 0.0055 0.0294 1.0000 0.0785 0.0167 0.0055 0.1230 IMF6 0.0054 0.0029 0.0053 0.0185 0.0785 1.0000 0.1167 0.0217 0.0620 IMF7 0.0038 0.0016 0.0018 0.0051 0.0167 0.1167 1.0000 0.0567 0.0411 IMF8 0.0023 0.0009 0.0010 0.0022 0.0055 0.0217 0.0567 1.0000 0.0240 L 1.0000 0.7302 0.3828 0.1758 0.1230 0.0620 0.0411 0.0240 1.0000

[0074] Table 2

[0075] Step 104: Introduce sample entropy and reconstruct the load sequence according to the sample entropy of each subsequence

[0076] First, calculate the sample entropy of each IMF according to formula (5). Then, based on the sample entropy results, reorganize the IMFs into two sequences. Specifically: calculate the sample entropy of all IMFs, use the quartiles of the sample entropy as the threshold for distinguishing the reorganized sequences. Sub - sequences with sample entropy lower than the threshold are classified into the low - complexity group F1, which usually contains more regular and periodic signal components. Sub - sequences with sample entropy higher than the threshold are classified into the high - complexity group F2, which usually contains more complex or noisy components. Evaluate the effectiveness of the reorganization by the correlation coefficients between the reorganized sequences F1, F2, and L. In the embodiment, there are 8 sub - sequences in total from IMF1 to IMF8. Figure 4 The images of the sample entropy of each IMF are shown, and the reorganized sequences are as Figure 5 shown. Denote the reorganized sequences as F1 and F2. The correlation coefficients after reconstruction are shown in Table 3. It can be seen that each sequence after reorganization not only retains the important feature information before reconstruction, but also there is almost no overlapping information between sequences, and the reorganization effect is good.

[0077] F1 F2 L F1 1.0000 0.0126 0.9794 F2 0.0126 1.0000 0.2139 L 0.9794 0.2139 1.0000

[0078] Table 3

[0079] Step 105: Perform NLM filtering on the high - complexity sequence

[0080] Further denoise the reorganized high - complexity group F2 through the non - local means filtering algorithm (NLM). The calculation process is shown in formulas (6) - (10). In the embodiment, the noisy signal u(t) is L, v(t) is the filtered signal to be solved, and n(t) is the noise interference signal.

[0081] Step 106: Input the original power load time series and the power load series reconstructed using this method into the same model for effect evaluation

[0082] Select a prediction model to predict the original power load time series and the power load time series reconstructed using this method in the same environment, and evaluate through four indicators: goodness of fit (R 2 ), mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) to verify the effectiveness of this method for predicting subsequent power load series. In the embodiment, the GRU model is selected for power load prediction. Table 3 shows the index changes when the data is input into the GRU model before and after being processed by this method. It can be seen that all indicators are effectively improved. The image between the output predicted value and the true value of the power load series processed by this method through the GRU model is as Figure 6 shown.

[0083] <![CDATA[R 2 > MAE / (KW) RMSE / (KW) MAPE / (%) Before treatment 0.9927 77.73 101.2 0.9602 After treatment 0.9997 43.07 56.29 0.5186

[0084] Table 4

[0085] According to an example of the present invention, the dataset has a total of 17,521 sample points, which are divided into a training set, a validation set, and a test set at 64%, 16%, and 20% respectively. When training, the number of epochs used is 50, the learning rate is 0.001, the batch size is selected as 128, and the deep learning framework used is TensorFlow. The server configuration used for training is as follows: 2 Intel Xeon E5-2698 v4 CPUs, 2 Nvidia RTX3090 GPUs, with a video memory of 24GB, a total machine memory of 128GB, and storage of 1TB SSD + 4TB HDD. It can be seen from the example that after the power load sequence is improved by the method proposed in the present invention, it is beneficial to improve the accuracy of subsequent power load prediction.

[0086] The above is only a preferred embodiment of the present invention, and it is not intended to limit the present invention in any other form. Any modification or equivalent change made according to the technical essence of the present invention still falls within the scope of the present invention claimed.

Claims

1. A GWO-VMD-NLM time series analysis method for improving the accuracy of power load forecasting, characterized in that: The specific steps are as follows: Step 101: Select the power load time series data to be processed and perform necessary preprocessing on the data; Step 102: Introduce the Akaike information criterion into the grey wolf optimization algorithm to automatically optimize and determine the key parameters of VMD: the secondary penalty factor α and the number of modes K. Use the optimized VMD algorithm to decompose the selected power load time series data into multiple subsequences; Step 103: Calculate the correlation coefficients between the decomposed subsequences and between each subsequence and the original load sequence to evaluate the effectiveness of the decomposition; Step 104: Introduce sample entropy. Recombine the load sequence into a low-complexity group F1 and a high-complexity group F2 according to the sample entropy of each subsequence, and evaluate the effectiveness of the reconstruction by the correlation coefficients between the recombined sequences and between each recombined sequence and the original load sequence; Step 105: Perform non-local means filtering on the high-complexity group to further remove the noise of the power load sequence while retaining important features; Step 106: Input the power load time series before and after being processed by this method into the same model for load prediction.

2. The GWO-VMD-NLM time series analysis method for improving the accuracy of electric power load forecasting according to claim 1, characterized in that: The specific steps of Step 101 are as follows; Process the missing values in the power load time series. Considering the non-linearity of the power load time series, the second-order polynomial interpolation method is used to fill in the missing values. Specifically, if the value of x i is missing or abnormal, then the first two sample points x i-1 , x i-2 and the last two sample points x i+1 and x i+2 are used to substitute into formula (1) to solve for the missing value of x i ; 3. The GWO-VMD-NLM time series analysis method for improving the accuracy of electric load forecasting according to claim 1, characterized in that: The specific steps of Step 102 are as follows; Step 102 first initializes the parameters of GWO and VMD, and uses the initialized parameters α and K to perform VMD decomposition on the preprocessed power load sequence to obtain a set of power load subsequences; then, using the Akaike information criterion, establish the objective function of GWO, and the expression is calculated as follows: Among them, <α, K> are the secondary penalty factor and the number of mode decompositions of VMD to be optimized, n is the number of sample points of the load sequence, and RSS is shown in formula (3): where y i and are the i-th values of the original load sequence and the reconstructed load sequence respectively, is calculated as shown in Equation (4) where, u k (t) is the k-th modal component obtained by VMD decomposition; Automatically optimize and update α and K through the objective function of GWO and continuously iterate until the best α and K are obtained. Determine them as the final VMD parameters and perform VMD decomposition on L to obtain a series of power load subsequences.

4. The GWO-VMD-NLM time series analysis method for improving the accuracy of electric power load forecasting according to claim 1, wherein: The specific steps of Step 103 are as follows; Calculate the correlation coefficient matrix between each IMF of the subsequences. If the absolute values of all correlation coefficients are less than 0.25, it proves that the decomposition effect is good.

5. The GWO-VMD-NLM time series analysis method for improving the accuracy of electric power load forecasting according to claim 1, wherein: The specific steps of Step 104 are as follows; First, calculate the sample entropy of each IMF. The sample entropy is defined as: Among them, B m (r) is the probability that two sequences match m points under the similarity tolerance r, while A m (r) is the probability that two sequences match m + 1 points. Then, based on the sample entropy results, the IMFs are recombined into two sequences, specifically as follows: Calculate the sample entropy of all IMFs, use the quartiles of the sample entropy as the threshold for distinguishing the recombined sequences, classify the subsequences with sample entropy lower than the threshold into the low-complexity group F1, which usually contains more regular and periodic signal components, and classify the subsequences with sample entropy higher than the threshold into the high-complexity group F2, which usually contains more complex or noisy components. Evaluate the effectiveness of the recombination by the correlation coefficients between the recombined sequences F1 and F2 and between each of them and L.

6. The GWO-VMD-NLM time series analysis method for improving the accuracy of electric load forecasting according to claim 1, wherein: The specific steps of Step 105 are as follows; Further denoise the recombined high-complexity group F2 through the non-local means filtering algorithm. The principle of NLM is as follows: u(t)v(t)+n(t) (6) Where u(t) is the noisy signal, v(t) is the original signal, n(t) is the noise interference signal, and the average estimated value of sample x is obtained by calculating the weighted set of different points t in the adjacent region: Among them, w(x, t) is the weight, v(x) is the output at the sample point x, z(x) is the accumulation of weights within the search domain, and the weight calculation is shown in Formulas (9) and (10): Among them, x and t are sample points, d represents the Gaussian weighted Euclidean distance, N(x) is the neighborhood block centered on x, and λ is a parameter affecting the smoothness of the filtered signal.

7. The GWO-VMD-NLM time series analysis method for improving the accuracy of electric power load forecasting according to claim 1, characterized in that: The specific steps of Step 106 are as follows; The selected prediction model will be used to predict the original power load time series and the power load time series processed by this method in the same environment, and the effectiveness of this method for subsequent power load series prediction will be verified by evaluating four indicators: the coefficient of determination R 2 square, mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE).

Images