Short-term power load prediction method based on modal Fourier decomposition and improved Autoformer

Abstract

The invention relates to a short-term power load prediction method based on modal Fourier decomposition and an improved Autoformer, and belongs to the technical field of power. The method comprises three parts, namely an influence factor correlation analysis module, a modal Fourier joint decomposition module and an integrated module for improving Autoformer. The short-term power load prediction method provided by the invention has relatively high prediction precision and relatively strong robustness under the condition that the load presents complex nonlinear characteristics, noise interference and obvious seasonal fluctuation. According to the method, a modal Fourier joint decomposition module is used for performing load decomposition test, robustness test, one-week test and one-day test, experimental results show that the modal Fourier joint decomposition module significantly reduces the complexity of load components, and compared with a traditional model, the method provided by the invention has better robustness, has higher stability in one-week and one-day tests, and has the highest prediction precision.

Description

Technical Field

[0001] The invention belongs to the technical field of electric power, and relates to a short-term electric power load forecasting method based on modal Fourier decomposition and improved Autoformer. Background Art

[0002] Accurate short-term power load forecasting is crucial to the stability and economic operation of the power grid. However, power load is affected by external factors such as user behavior, weather, and holidays, among which meteorological factors have a greater impact. In recent years, the frequency and intensity of extreme weather events have increased, and people's demand for cooling and heating has increased sharply, causing a surge in power load demand, resulting in strong power load fluctuations and a lot of noise, which has brought huge challenges to short-term power load forecasting.

[0003] Existing power load forecasting methods mainly include traditional statistical methods, machine learning methods, neural network methods and combined model methods. Traditional statistical methods, such as autoregressive model (AR), autoregressive integrated moving average model (ARIMA) and exponential smoothing method, have simple structure and fast training speed, but they are difficult to capture nonlinear trends in power load and are sensitive to noise. Machine learning methods include random forest, gradient boosting tree and support vector machine. Compared with traditional statistical methods, machine learning methods can effectively handle nonlinear features in data and show stronger generalization ability, but are usually suitable for scenarios with low complexity of load forecasting data. Neural network methods, such as long short-term memory network (LSTM), temporal convolutional network (TCN), transformer and autoformer models. LSTM processes time series recursively, and has the problem of gradient explosion; TCN has good time series feature capture capabilities, but insufficient ability to capture long-term features; Transformer has strong nonlinear mapping capabilities, but is less efficient in processing long-term time series; Autoformer is stronger in extracting features from sequences than Transformer, but has the problem of missing convolution, resulting in insufficient ability to capture short-term features.

[0004] Many studies have proposed combined model methods, such as: a combined prediction model combining K-means and extreme learning machine, which clusters user behavior features through K-means, reduces the dimension and complexity of feature data, and uses extreme learning machine for prediction; VMD-VAE-LSTM hybrid load prediction model, which decomposes data into multiple subsequences, extracts features from the decomposed sequences by VAE, and finally uses LSTM for prediction; a model combining Complete En-semble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), sample entropy and Transformer, which decomposes the original load sequence and combines Transformer for prediction. The above methods do not conduct in-depth analysis of the correlation between load components and do not fully consider the influencing factors, resulting in too high a dimension of load components and increasing the difficulty of calculation. Due to the scale limitation of the model for feature extraction, the extraction of local features is not comprehensive, which in turn affects the robustness and prediction accuracy of the prediction model. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide a short-term power load forecasting method based on modal Fourier decomposition and improved Autoformer (Mode Fourier-Improved Autoformer Joint Prediction Method, MF-IAFP) to solve the problems of insufficient robustness and insufficient prediction accuracy of existing short-term power load forecasting methods.

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

[0007] A short-term power load forecasting method based on modal Fourier decomposition and improved Autoformer, the method comprising the following steps:

[0008] S1: Use the Spearman correlation coefficient method to analyze the correlation between load and other influencing factors and screen out key influencing factors;

[0009] S2: Combine Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) with Fast Fourier Transform (FFT) and propose Mode Fourier Joint Analysis (MFJA) to decompose the original power load.

[0010] S3: Establish a load component data set of key influencing factors and load components;

[0011] S4: Introducing the Temporal Convolutional Network (TCN) to improve the autocorrelation model in Autoformer;

[0012] S5: Establish an integrated IAutoformer model for each load component data set and perform prediction;

[0013] S6: Fusion the prediction values ​​of each model and output the final prediction result.

[0014] Further, the S1 comprises the following steps:

[0015] 1) Collect power load data through the power load monitoring platform;

[0016] 2) Obtain meteorological factor data such as temperature, wind speed, rainfall and humidity through the meteorological platform;

[0017] 3) Process missing values, outliers, and standardize and normalize the collected data;

[0018] 4) Use independent codes to code working days and hours;

[0019] 5) Use the Spearman correlation coefficient method to conduct correlation analysis on load data and various influencing factors, and screen out influencing factors related to load.

[0020] Furthermore, the modal Fourier joint analysis method comprises the following steps:

[0021] 1) Input the original power load sequence into the fully adaptive noise ensemble empirical mode decomposition model to obtain load components of different frequencies;

[0022] 2) Use fast Fourier transform to obtain the main frequency of each component;

[0023] 3) Setting the frequency difference threshold Δf;

[0024] 4) Calculate the frequency difference between the two components. If the difference is less than Δf, merge them for optimization;

[0025] 5) Finally, output the new load component.

[0026] Further, the specific steps of S3 are as follows:

[0027] According to the screened key influencing factors and the output load components, load component data sets of each load component are established respectively.

[0028] Further, the S4 comprises the following steps:

[0029] 1) After linear transformation of the query matrix Q and the key matrix K, the frequency domain representation is obtained through FFT, and then the frequency domain feature Z is obtained by splicing f ;

[0030] 2) Perform a linear transformation on the value matrix V to obtain V1;

[0031] 3) The frequency domain feature Z f Through the inverse Fourier transform (IFFT) back to the time domain to get the time domain feature Z t ;

[0032] 4) Z t , V1 is subjected to delayed aggregation to obtain the time domain feature Z after delayed aggregation g , Z c is the result after splicing;

[0033] 5) The concatenated features are extracted at multiple scales through the TCN structure and output through the linear layer.

[0034] Further, the establishment of the integrated IAutoformer model includes the following steps:

[0035] 1) Establish IAutoformer models respectively according to the established load component data sets;

[0036] 2) Divide the load component data set into training set and test set;

[0037] 3) Train each model and adjust parameters.

[0038] Furthermore, the prediction values ​​of each model are integrated to output the final prediction result, which includes the following steps:

[0039] According to the established integrated IAutoformer model, the divided test sets are input into each model, and the predicted output values ​​of each model are fused to obtain the final output prediction result.

[0040] The beneficial effects of the present invention are:

[0041] 1. Use the influencing factor correlation analysis module to conduct correlation analysis on external factors such as temperature, humidity, wind speed, time and short-term power load, screen out key influencing factors, and significantly reduce the calculation complexity of model prediction.

[0042] 2. Compared with the modal components obtained by CEEMDAN decomposition, the modal components processed by the modal Fourier joint decomposition module significantly reduce the modal complexity and effectively reduce the volatility of the mid- and low-frequency components.

[0043] 3. Compared with other short-term power load forecasting models, the method proposed in the present invention shows the highest prediction accuracy on different data sets. At the same time, the method proposed in the present invention is superior to other traditional models in terms of anti-interference ability and stability, and has the best robustness.

[0044] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0046] Figure 1 The structure and data flow diagram of TCN provided by the embodiment of the present invention;

[0047] Figure 2 An improved autocorrelation module structure provided by an embodiment of the present invention;

[0048] Figure 3 The overall architecture diagram of the MF-IAFP model provided by the embodiment of the present invention;

[0049] Figure 4 A heat map of the correlation coefficient of the impact factors provided by the embodiment of the present invention;

[0050] Figure 5 MFJA test result diagram provided by an embodiment of the present invention;

[0051] Figure 6 A one-week test result chart of each data set provided in an embodiment of the present invention;

[0052] Figure 7 A box plot of the one-week test error of each data set provided by the embodiment of the present invention;

[0053] Figure 8 A one-day test result diagram of each data set provided in an embodiment of the present invention;

[0054] Fig. 9 A box plot of the one-day test error of each data set provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0055] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0056] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0057] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0058] This example of the present invention provides a short-term power load forecasting method based on modal Fourier decomposition and improved Autoformer. The method mainly includes an influencing factor correlation analysis module, a modal Fourier joint decomposition module and an integrated module of the improved Autoformer to complete short-term power load forecasting.

[0059] This embodiment first describes the basic principles and execution processes of the above three parts as a whole, and then describes the specific steps of each part as follows:

[0060] Step 1: Conduct correlation analysis of influencing factors

[0061] The power load data, temperature, rainfall, humidity, wind speed and other meteorological factors are collected through the power real-time monitoring platform and the meteorological collection platform, and the collected data is preprocessed, including abnormal data detection, missing value filling, data standardization and normalization, so as to improve the accuracy of the correlation calculation between the data. The Spearman correlation coefficient method is used to analyze the correlation between the load and various external influencing factors, and the influencing factors with high correlation are screened as the input features of the subsequent model.

[0062] Step 2: Decomposition of loads using modal Fourier joint analysis

[0063] The load sequence is decomposed by CEEMDAN to obtain several stationary subsequences with different frequency characteristics (IMF1, IMF2, ..., IMFn). Then, the main frequency of each load component is extracted by FFT, and the coupling processing of the modes with similar frequencies is performed according to the set frequency difference threshold, and finally the processed load components are output; the execution process is as follows:

[0064] 1) Input the original power load sequence x(t), use the CEEMDAN algorithm to decompose x(t) and obtain a series of IMF components IMFi (where i = 1, 2, ..., n).

[0065] 2) Apply FFT to each IMF component to obtain the spectrum. i (f) indicates that each IMF i The main frequency f i .

[0066] 3) Setting the frequency difference threshold Δf, if the difference in the main frequencies of the two IMF components is less than the set threshold, it is considered that they can be fused.

[0067] 4) Fusion rules: For adjacent IMFs i With IMF i+1 , if |f i -f i+1 |<Δf, they are fused and the fused new IMF component is recorded as FIMF j (where j = 1, 2, ..., m), the calculation formula is:

[0068] FIMF j =IMF i +IMF i+1

[0069] 5) The fusion component FIMF j (where j = 1, 2, ..., m) is used as the final result after the original power load decomposition and component fusion, and is input into the data set of the fusion influencing factors.

[0070] Step 3: Improve the integration module of Autoformer

[0071] Based on the influencing factors selected by the influencing factor correlation analysis module and the load components output by the modal Fourier joint decomposition module, m load component data sets are constructed. For each load component data set, an independent improved Autoformer prediction model is established, and each load component is predicted. Finally, the prediction results of each model are reconstructed and integrated to obtain the overall load prediction result. The execution process is:

[0072] 1) Build TCN network

[0073] The structure of TCN is Figure 1 As shown in the figure, it consists of two submodules connected in series. Each submodule contains components such as dilated causal convolution layer, Dropout layer, ReLU activation function and weight normalization (WeightNorm). The dilation factor of the first layer of dilated causal convolution is set to 1, which is mainly used to extract short-term dependency features; the dilation factor of the second layer is set to 2 to capture longer-term dependency features.

[0074] 2) Improve the autocorrelation module

[0075] The improved autocorrelation module structure is as follows: Figure 2 As shown in Figure 1, the output of the original autocorrelation module is subjected to a time convolution operation to extract the complex relationship between local load and load, and between load and influencing factors. The data processing steps of the improved autocorrelation module are as follows:

[0076] After linear transformation of the query matrix Q and the key matrix K, the frequency domain representation Q1 of the query matrix and the frequency domain representation K1 of the key matrix are obtained by FFT respectively;

[0077]

[0078] Frequency domain feature Z f It is expressed as:

[0079]

[0080] In the formula Represents frequency domain multiplication.

[0081] Z f Perform an inverse Fourier transform (IFFT) to obtain the time domain feature Z t for:

[0082] Z t =IFFT(Z f )

[0083] After linear transformation of the value matrix V, V1 is obtained, and Z t Perform delayed aggregation with V1 to obtain the aggregate feature Z g ;

[0084] V1=Linear(V)

[0085] Z g =TimeDelayAgg(Z t ,V1)

[0086] And for Z g After splicing, we get the splicing feature Z c .

[0087] Z c =Concat(Z g )

[0088] Z c After multi-scale feature extraction by TCN and output through the linear layer, the output result Z of the improved autocorrelation module is obtained. out .

[0089] Z out =TCN(Z c )

[0090] 3) Establishment of MF-IAFP model

[0091] Figure 3 This is the overall architecture diagram of the MF-IAFP model, including influencing factor analysis, modal Fourier joint analysis method, and the establishment of an integrated short-term power load forecasting model. The short-term power load forecasting model is integrated to build a load component data set. For each load component data set, an independent improved Autoformer forecasting model is established, and each load component is predicted. Finally, the prediction results of each model are reconstructed and integrated to obtain the overall load forecasting result.

[0092] Based on the overview of the above three parts, this embodiment provides a short-term power load forecasting method based on modal Fourier decomposition and improved Autoformer, including the following steps:

[0093] Step 1: Conduct correlation analysis of influencing factors

[0094] Step 1.1: Collect power load data, temperature, rainfall, humidity, wind speed and other meteorological factors through the power real-time monitoring platform and meteorological collection platform.

[0095] Step 1.2: Preprocess the collected data, including abnormal data detection, missing value filling, data standardization and normalization, so as to improve the accuracy of correlation calculation between data.

[0096] Step 1.3: Use the Spearman correlation coefficient method to analyze the correlation between the load and various external influencing factors, and screen the influencing factors with high correlation as the input features of the subsequent model.

[0097] This example collects data sets from three different regions as research objects, among which data set 1 is a load data set actually collected in a certain place in the southwest, with a time interval of 15 minutes; data set 2 is a public data set in a certain place in central China, with a time interval of 15 minutes; data set 3 is a public data set in a certain place in Australia, with a time interval of 0.5 hours; taking data set 1 as an example, the influencing factors are screened, the modal Fourier joint analysis method is tested, and the robustness test is performed. At the same time, in order to verify the stability and prediction accuracy of MF-IAFP and other models on different time scales and different data sets, this paper conducts a one-week prediction test and a one-day prediction test on the above data sets. The training rounds (epochs) of each model in the experiment are 100, the optimizer selects Adam, the learning rate is set to 0.0005, and the dropout rate is 0.3 to ensure that the performance of each model is compared under the same experimental conditions.

[0098] The Spearman correlation coefficient was used to analyze the correlation between each influencing factor and the load. The absolute value of the correlation coefficient was between 0.8 and 1, which was a very strong correlation; between 0.6 and 0.8, which was a strong correlation; between 0.4 and 0.6, which was a moderate correlation; between 0.2 and 0.4, which was a weak correlation; and between 0 and 0.2, which was no correlation. Taking data set 1 as an example, the correlation coefficient is as follows: Figure 4 As shown in the figure, the correlation coefficients of Temperature, Rainfall, Month, Hour and Load are 0.67, 0.36, 0.25 and 0.42 respectively, so they are used as input features to establish the short-term power load forecasting model.

[0099] Step 2: Decompose the power load using the modal Fourier joint analysis method

[0100] Step 2.1: Input the pre-processed power load data into CEEMDAN for decomposition to obtain power load components of different frequencies;

[0101] Step 2.2: Use FFT to extract the main frequency of each load component;

[0102] Step 2.3: By setting Δf, components with smaller frequency differences are selected for fusion optimization, the complexity of the modal components is reduced, and the final decomposition results are output.

[0103] The load time series of Dataset 1 for 15 consecutive days was selected for analysis. Figure 5The results of the MFJA test are shown. This method uses CEEMDAN to decompose the original load time series data into 8 intrinsic modal components (IMFs). IMF1 and IMF2 are high-frequency components, which mainly capture the noise and short-term changes in the load series, reflecting the strong random fluctuation characteristics in the load; IMF3 to IMF5 are medium-frequency components, which reveal the daily fluctuation law in the load; IMF6 to IMF8 are low-frequency components, reflecting the long-term trend and seasonal changes in the load series. Secondly, the main frequency of each modal component is analyzed by FFT as shown in Table 1. The coupling components in the load components are processed by setting Δf, where IMF4 and IMF5 are merged into FIMF4, and IMF6 and IMF7 are merged into FIMF5. Compared with the modal components decomposed by CEEMDAN, the modal components processed by MFJA significantly reduce the modal complexity and effectively reduce the volatility of the medium and low frequency components.

[0104] Table 1

[0105] Quantity IMF1 IMF2 IMF3 IMF4 IMF5 IMF6 IMF7 IMF8 frequency 0.26041 0.125 0.03125 0.01041 0.01041 0.00138 0.00138 0.00001

[0106] Step 3: Build the MF-IAFP model

[0107] Step 3.1: Construct a load component data set based on the influencing factors selected by the influencing factor correlation analysis module and the load components output by the modal Fourier joint decomposition module;

[0108] Step 3.2: Establish an IAutoformer prediction model for each load component data set;

[0109] Step 3.3: Conduct short-term power load forecasting;

[0110] Step 3.4: Fusion the predicted values ​​of each model output value to obtain the final prediction result.

[0111] One-week test: The MF-IAFP model proposed in this paper was compared and analyzed with LSTM, CNN, Transformer, Autoformer, IAutoformer and MF-Autoformer on different data sets. Figure 6The weekly prediction results of each model on different data sets are shown. On data set 1, the prediction results of CNN and Transformer are poor, and the prediction curves of other models are relatively close to the true value. By zooming in on a trough, it can be found that the MF-IAFP model performs well in capturing subtle load fluctuations, and the prediction results are closest to the true value. On data set 2, CNN, Autoformer, Transformer, and IAutoformer have obvious deviations in the prediction of peaks and troughs. In contrast, LSTM, MF-Autoformer, and MF-IAFP have better prediction results. In particular, when zooming in on a peak, the prediction curve of MF-IAFP is highly consistent with the true value, showing strong prediction ability. On data set 3, the prediction results of Autoformer at a trough deviate significantly from the true value, while the prediction results of other models are relatively good. When zooming in on a bimodal area, the prediction value of MF-IAFP is closest to the true value, further verifying its superiority in complex fluctuation patterns. The MF-IAFP model shows stronger prediction ability on all data sets, especially in capturing subtle fluctuations in load time series.

[0112] Figure 7 The box plots of the one-week test errors of each data set are shown in Figure 1. As can be seen from the figure, the box height of the MF-IAFP model in each data set is the lowest, indicating that the distribution of its prediction errors is the most concentrated.

[0113] Table 2 shows the results of the one-week test indicators of each model on different data sets. In the test of data set 1, MF-IAFP has a MAPE of 0.552, a RMSE of 57.902, a MAE of 45.267, and an R 2 The MAPE, RMSE, MAE and R are 0.9978, which are 4%, 5.37%, 3.23% and 0.03% higher than the other optimal models respectively. In the test of dataset 2, the MAPE, RMSE, MAE and R are 0.664, 577.587, 434.113 and 0.03% higher than the other optimal models respectively. 2 is 0.9971, which is 41.91%, 39.03%, 42.87%, and 0.49% higher than the other optimal models. In dataset 3, MF-IAFP has a MAPE of 0.424, a RMSE of 1919.863, a MAE of 1543.969, and an R 2 is 0.9837, which is an improvement of 11.48%, 10.85%, 11.65%, and 0.43% respectively compared with the other optimal models.

[0114] Table 2

[0115]

[0116] One-day test: The present invention conducts one-day test comparison and analysis on each model on different data sets. The results are as follows: Figure 8 As shown. In Dataset 1, in the 0-4 hour interval, the prediction results of MF-IAFP are the most accurate, while other models have large deviations; in the 8-12 hour interval, the prediction curve of MF-IAFP closely follows the true value fluctuation, while the prediction curves of other models are relatively smooth and have large deviations. In Dataset 2, the prediction curves of Transformer, Autoformer, and MF-Autoformer deviate seriously from the true value in some time periods, and the prediction effects of LSTM, CNN, and MF-IAFP are relatively good, among which MF-IAFP has the best prediction effect. In Dataset 3, the prediction effects of each model are generally good. After zooming in on the 12-14 hour interval, it is found that the prediction results of MF-IAFP are the most accurate.

[0117] Fig. 9 The error box plots of the one-day test for each data set are shown. Consistent with the one-week test results, the box height of the MF-IAFP model is the smallest, indicating that its error value distribution is the most concentrated.

[0118] Table 3 shows the one-day test results of each model on three datasets. In the test of dataset 1, MF-IAFP has a MAPE of 0.439, a RMSE of 46.360, a MAE of 36.501, and an R 2 is 0.9981, which is 33.58%, 34.12%, 34.73% and 0.25% higher than the other optimal models. In Dataset 2, MF-IAFP has a MAPE of 0.807, a RMSE of 559.041, a MAE of 491.804, and R 2 is 0.9959, which is 13.87%, 35.93%, 17.47% and 0.58% higher than the other optimal models. From Dataset 3, MF-IAFP has a MAPE of 0.296, a RMSE of 1406.81, a MAE of 1088.799, and R 2 It reaches 0.9927. Compared with other best-performing models, MF-IAFP improves 36.09%, 34.22%, 35.49% and 0.9% in various indicators respectively, fully demonstrating its excellent prediction accuracy.

[0119] Table 3

[0120]

[0121] Robustness test: Based on data set 1, the present invention conducts robustness tests on each model. Three conditions are considered in the test: common random noise (Gaussian noise), extreme interference (uniform noise) and ideal conditions (no noise). By comparing these three conditions, the robustness and adaptability of each model can be comprehensively evaluated. The test results are shown in Table 4. Under noise-free conditions, the MAPE of MF-IAFP is 0.296, the RMSE is 1406.818, the MAE is 1088.799, and the R 2 Under Gaussian noise interference, the MAPE of MF-IAFP is 0.409, the RMSE is 1828.211, the MAE is 1495.68, and the R 2 The value is 0.9877. Under uniform noise interference conditions, the MAPE of MF-IAFP is 0.320, the RMSE is 1505.592, the MAE is 1178.789, and the R 2 is 0.9917. Comparing the differences in MAPE, RMSE and MAE of each model under Gaussian noise, uniform noise and no noise conditions, the results are shown in Table 5. MF-IAFP has the smallest value in all the differences: the differences in MAPE are 0.113 and 0.024, the differences in RMSE are 421.393 and 98.774, and the differences in MAE are 406.881 and 89.99. These results show that the MF-IAFP model has strong robustness and can maintain excellent prediction performance in complex noise environments.

[0122] Table 4

[0123]

[0124] Table 5

[0125]

[0126]

[0127] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. 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 solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A short-term power load forecasting method based on modal Fourier decomposition and improved Autoformer, characterized by: The method comprises the following steps: S1: Use the Spearman correlation coefficient method to analyze the correlation between load and other influencing factors and screen out key influencing factors; S2: Combine Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) with Fast Fourier Transform (FFT) and propose Mode Fourier Joint Analysis (MFJA) to decompose the original power load. S3: Establish a load component data set of key influencing factors and load components; S4: Introducing the Temporal Convolutional Network (TCN) to improve the autocorrelation model in Autoformer; S5: Establish an integrated IAutoformer model for each load component data set and perform prediction; S6: Fusion the prediction values ​​of each model and output the final prediction result.

2. According to claim 1, a short-term power load forecasting method based on modal Fourier decomposition and improved Autoformer is characterized by: The S1 comprises the following steps: 1) Collect power load data through the power load monitoring platform; 2) Obtain meteorological factor data such as temperature, wind speed, rainfall and humidity through the meteorological platform; 3) Process missing values, outliers, and standardize and normalize the collected data; 4) Use independent codes to code working days and hours; 5) Use the Spearman correlation coefficient method to conduct correlation analysis on load data and various influencing factors, and screen out influencing factors related to load.

3. According to claim 1, a short-term power load forecasting method based on modal Fourier decomposition and improved Autoformer is characterized by: The modal Fourier joint analysis method comprises the following steps: 1) Input the original power load sequence into the fully adaptive noise ensemble empirical mode decomposition model to obtain load components of different frequencies; 2) Use fast Fourier transform to obtain the main frequency of each component; 3) Setting the frequency difference threshold Δf; 4) Calculate the frequency difference between the two components. If the difference is less than Δf, merge them for optimization; 5) Finally, output the new load component.

4. According to claim 2, a short-term power load forecasting method based on modal Fourier decomposition and improved Autoformer is characterized by: The specific steps of S3 are as follows: According to the screened key influencing factors and the output load components, load component data sets of each load component are established respectively.

5. The short-term power load forecasting method based on modal Fourier decomposition and improved Autoformer according to claim 1 is characterized by: The S4 comprises the following steps: 1) After linear transformation of the query matrix Q and the key matrix K, the frequency domain representation is obtained through FFT, and then the frequency domain feature Z is obtained by splicing f ; 2) Perform a linear transformation on the value matrix V to obtain V1; 3) The frequency domain feature Z f Through the inverse Fourier transform (IFFT) back to the time domain to get the time domain feature Z t ; 4) Z t , V1 is subjected to delayed aggregation to obtain the time domain feature Z after delayed aggregation g , Z c is the result after splicing; 5) The concatenated features are extracted at multiple scales through the TCN structure and output through the linear layer.

6. According to claim 4, a short-term power load forecasting method based on modal Fourier decomposition and improved Autoformer is characterized by: The establishment of the integrated IAutoformer model comprises the following steps: 1) Establish IAutoformer models respectively according to the established load component data sets; 2) Divide the load component data set into training set and test set; 3) Train each model and adjust parameters.

7. The short-term power load forecasting method based on modal Fourier decomposition and improved Autoformer according to claim 6 is characterized by: Fusion of the prediction values ​​of each model and output of the final prediction result includes the following steps: According to the established integrated IAutoformer model, the divided test sets are input into each model, and the predicted output values ​​of each model are fused to obtain the final output prediction result.

Images