Wind power prediction method, system and equipment based on secondary decomposition and reconstruction and medium

Abstract

The invention relates to the technical field of wind power prediction, and discloses a wind power prediction method, system and device based on secondary decomposition and reconstruction and a medium, and the method comprises the steps: obtaining first historical power data of a wind power station; performing complete empirical mode decomposition on the first historical power data to obtain a plurality of intrinsic mode functions of different frequencies corresponding to the first historical power data; clustering the intrinsic mode functions based on the frequency characteristics to obtain a plurality of clusters; performing secondary decomposition on each intrinsic mode function based on the frequency characteristics of each cluster to obtain secondary decomposition sub-signals corresponding to each cluster; inputting each secondary decomposition sub-signal into a pre-constructed preset prediction model for power prediction to obtain a sub-power prediction result corresponding to each secondary decomposition sub-signal; and fusing the sub-power prediction results to obtain a wind power prediction result. According to the method, the complex fluctuation mode in the wind power time sequence can be effectively extracted, and the prediction result is more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power prediction, and in particular to a wind power prediction method, system, equipment and medium based on secondary decomposition and reconstruction. Background Art

[0002] As the global energy mix shifts toward low-carbon, clean energy, wind power, as a key component of renewable energy, is gaining widespread adoption due to its environmentally friendly and sustainable nature. However, wind power output is highly dependent on meteorological conditions, such as wind speed, direction, temperature, and air pressure. The randomness and volatility of these factors pose significant challenges to wind power forecasting. Accurate wind power forecasting is crucial for grid stability, economic dispatch, and the operation of the electricity market. Large forecast errors can lead to grid frequency fluctuations, insufficient reserve capacity, or wind curtailment and power rationing, impacting the safety and affordability of the power system.

[0003] In existing technologies, wind power forecasting is primarily based on historical meteorological data and wind farm power data, using machine learning or statistical methods to build prediction models. Traditional methods typically employ algorithms such as linear regression, support vector regression, and artificial neural networks to fit the mapping relationship between meteorological data and wind power, thereby predicting future power. However, these methods have the following limitations: ① Wind power is affected by the coupling of multiple meteorological factors, exhibiting non-stationary, nonlinear, and multi-timescale variations. Traditional linear or shallow nonlinear models struggle to capture these complex dynamic characteristics, limiting prediction accuracy. ② Wind farm power output is affected not only by local meteorological conditions but also by wind farm fluctuations in surrounding areas. Most existing methods only consider single-point meteorological data, ignoring the spatiotemporal correlations within wind farms and across regions, which impacts the robustness of predictions. ③ Under conditions of strong winds, turbulence, or extreme temperatures, wind power can fluctuate dramatically. Traditional models, lacking the ability to quantify uncertainty, struggle to adapt to such abnormal scenarios. ④ Historical data may contain missing data, noise, or biases, and traditional methods rely on manual feature extraction, making it difficult to adaptively learn from the valid information in the data, thus affecting the model's generalization performance.

[0004] To overcome these challenges, recent research has attempted to improve forecast accuracy using deep learning or hybrid models. However, key technical challenges remain: how to more effectively model the multi-scale fluctuations of wind power and integrate multi-source data to enhance forecast accuracy and robustness. Therefore, a more advanced wind power forecasting method is urgently needed that can fully exploit the spatiotemporal evolution of wind power and adaptively learn its complex nonlinear relationships with meteorological factors to improve forecast accuracy and support the safe and stable operation of the power grid. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides a wind power prediction method, system, equipment and medium based on secondary decomposition and reconstruction to solve the problem that the method of fitting the linear or nonlinear relationship between historical meteorological data and historical power data does not take into account the complex fluctuation pattern of wind power, resulting in inaccurate power prediction.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, the present invention provides a wind power prediction method based on secondary decomposition and reconstruction, comprising:

[0009] Obtaining the first historical power data of the wind farm;

[0010] Performing complete empirical mode decomposition on the first historical power data to obtain a plurality of intrinsic mode functions of different frequencies corresponding to the first historical power data;

[0011] Clustering the intrinsic mode functions based on frequency characteristics to obtain multiple clusters;

[0012] Performing a secondary decomposition on each of the intrinsic mode functions based on the frequency characteristics of each of the clusters to obtain a secondary decomposition sub-signal corresponding to each of the clusters;

[0013] Inputting each of the secondary decomposition sub-signals into a pre-built preset prediction model to perform power prediction, and obtaining a sub-power prediction result corresponding to each of the secondary decomposition sub-signals;

[0014] The sub-power prediction results are integrated to obtain a wind power prediction result.

[0015] As a preferred solution of the wind power prediction method based on secondary decomposition and reconstruction described in the present invention, wherein: the clustering of each of the intrinsic mode functions based on frequency characteristics to obtain multiple clusters includes:

[0016] Extracting time-frequency features of each intrinsic mode function to obtain time-frequency features corresponding to each intrinsic mode function;

[0017] Clustering the intrinsic mode functions based on the similarity of the frequency characteristics of each of the intrinsic mode functions to obtain a first target cluster, a second target cluster, and a third target cluster;

[0018] The first target frequency domain in the first target cluster is larger than the second target frequency domain in the second target cluster, and the second target frequency domain in the second target cluster is larger than the third target frequency domain in the third target cluster.

[0019] The beneficial effect of this preferred technical solution is: using the clustering method to divide the intrinsic mode functions into clusters with similar dynamic characteristics, while retaining the time-frequency correlation within the cluster and capturing the potential coupling relationship between different wave modes.

[0020] As a preferred solution of the wind power prediction method based on secondary decomposition and reconstruction described in the present invention, the obtaining of the secondary decomposition sub-signals corresponding to each of the clusters includes:

[0021] Converting each of the intrinsic mode functions into a trajectory matrix using a sliding window of a preset window size;

[0022] Performing singular value decomposition on the trajectory matrix to obtain singular values and an orthogonal matrix of the trajectory matrix;

[0023] Sorting and screening the singular values to obtain target singular values;

[0024] Selecting a target singular vector based on the target singular value, and constructing a plurality of sub-target sequences based on the target singular value, the target singular vector and the orthogonal matrix;

[0025] Diagonal averaging is performed on each of the sub-target sequences to obtain the secondary decomposition sub-signal.

[0026] The beneficial effect of this preferred technical solution is that it can effectively extract complex fluctuation patterns in wind power time series.

[0027] As a preferred solution of the wind power prediction method based on secondary decomposition and reconstruction described in the present invention, the obtaining of the sub-power prediction results corresponding to each of the secondary decomposition sub-signals includes:

[0028] Inputting each of the secondary decomposition sub-signals into a pre-trained preset prediction model, and extracting the causal dependency and reverse time information of each of the secondary decomposition sub-signals based on the preset prediction model;

[0029] Reversing each of the reverse time information into reverse time information;

[0030] The sub-power prediction result is obtained based on the causal relationship and the flip time information.

[0031] As a preferred solution of the wind power prediction method based on secondary decomposition and reconstruction described in the present invention, wherein: the preset prediction model includes a cascaded bidirectional encoder module, a timing alignment module and a power output module;

[0032] The bidirectional encoder module includes a forward encoder and a reverse encoder connected in parallel, the forward encoder is used to extract the causal dependency of each of the secondary decomposition sub-signals, and the reverse encoder is used to extract reverse time information of each of the secondary decomposition sub-signals;

[0033] The timing alignment module is used to flip the reverse time information to obtain the flip time information;

[0034] The power output module is used to map the causal relationship and the reversal time information into the sub-power prediction result.

[0035] As a preferred solution of the wind power prediction method based on secondary decomposition and reconstruction described in the present invention, the fusing of the sub-power prediction results to obtain the wind power prediction result includes:

[0036] Obtaining the start time and end time of each sub-power prediction result;

[0037] Inputting the start time and the end time into an error prediction model to obtain a prediction error of each sub-power prediction result; wherein the error prediction model is trained based on the time information and timing dependency corresponding to the historical prediction errors of the preset prediction model, and during the error prediction model training process, the corresponding relationship between the moment changes of the historical prediction errors within the time information and the timing dependency is learned;

[0038] Based on the prediction error and each of the sub-power prediction results, a ridge regression method is used to obtain the wind power prediction result.

[0039] The beneficial effects of this preferred technical solution are: power prediction is performed based on local fluctuation characteristics and global interactions, and the prediction results are more accurate.

[0040] As a preferred solution of the wind power prediction method based on secondary decomposition and reconstruction described in the present invention, the training of the error prediction model includes:

[0041] Obtaining a historical sub-prediction result of the preset prediction model and second historical power data corresponding to the time series of the historical sub-prediction result; wherein the time of the second historical power is earlier than that of the first historical power data;

[0042] Obtaining a prediction error sequence with a timestamp based on the historical sub-prediction results and the second historical power data;

[0043] The prediction error sequence is input into a pre-built preset correction model for model training. During the model training process, the correspondence between the moment changes of the prediction error sequence in the time interval corresponding to the timestamp and the timing dependency is learned until the model converges to obtain the error prediction model.

[0044] In a second aspect, the present invention provides a wind power prediction system based on secondary decomposition and reconstruction, comprising:

[0045] A data acquisition module, configured to acquire first historical power data of a wind farm;

[0046] a first signal decomposition module, configured to perform complete empirical mode decomposition on the first historical power data to obtain a plurality of intrinsic mode functions of different frequencies corresponding to the first historical power data;

[0047] A clustering module, configured to cluster the intrinsic mode functions based on frequency characteristics to obtain a plurality of clusters;

[0048] A second signal decomposition module is used to perform secondary decomposition on each of the intrinsic mode functions based on the frequency characteristics of each of the clusters to obtain secondary decomposition sub-signals corresponding to each of the clusters;

[0049] A sub-power prediction module is used to input each of the secondary decomposition sub-signals into a pre-built preset prediction model to perform power prediction, and obtain a sub-power prediction result corresponding to each of the secondary decomposition sub-signals;

[0050] The power fusion module is used to fuse the sub-power prediction results to obtain a wind power prediction result.

[0051] In a third aspect, the present invention provides an electronic device comprising a memory and a processor; the memory is used to store computer-executable instructions, and the processor implements the steps of a wind power prediction method based on secondary decomposition and reconstruction when executing the computer-executable instructions.

[0052] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of a wind power prediction method based on secondary decomposition and reconstruction.

[0053] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention provides a wind power prediction method, system, device and medium based on secondary decomposition and reconstruction, which effectively retains the global interaction characteristics of sub-signals through a three-stage collaborative mechanism of complete empirical mode decomposition, clustering and secondary decomposition: performing complete empirical mode decomposition on the first historical power data to obtain multiple intrinsic mode functions, and using a clustering method to divide the intrinsic mode functions into clusters with similar dynamic characteristics, while retaining the time-frequency correlation within the cluster, capturing the potential coupling relationship between different fluctuation modes; performing secondary decomposition on each intrinsic mode function based on the frequency characteristics of each cluster, the secondary decomposition sub-signal retains the global interaction and local details of the first historical wind power in different frequency domains, and can effectively extract the complex fluctuation pattern in the wind power time series; using the secondary decomposition sub-signal to perform power prediction to obtain the power prediction result at the future time, the preset prediction model can extract the local fluctuation characteristics and global interaction of each intrinsic mode function by extracting the sequential causal relationship and reverse time information contained in each secondary decomposition sub-signal, so as to perform power prediction based on the local fluctuation characteristics and global interaction, and the prediction result is more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0055] Figure 1 This is a logical diagram of the overall process of a wind power prediction method based on secondary decomposition and reconstruction according to an embodiment of the present invention. DETAILED DESCRIPTION

[0056] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0057] Example 1, reference Figure 1 As an embodiment of the present invention, a wind power prediction method based on secondary decomposition and reconstruction is provided, such as Figure 1 The specific steps shown include:

[0058] S100: Acquire first historical power data of a wind farm;

[0059] S200: performing complete empirical mode decomposition on the first historical power data to obtain a plurality of intrinsic mode functions of different frequencies corresponding to the first historical power data;

[0060] S300: Clustering each intrinsic mode function based on frequency characteristics to obtain multiple clusters;

[0061] S400: performing secondary decomposition on each intrinsic mode function based on the frequency characteristics of each cluster to obtain a secondary decomposition sub-signal corresponding to each cluster;

[0062] S500: Inputting each secondary decomposition sub-signal into a pre-built preset prediction model to perform power prediction, and obtaining a sub-power prediction result corresponding to each secondary decomposition sub-signal;

[0063] S600: Fusing the sub-power prediction results to obtain a wind power prediction result.

[0064] It should be noted that wind power has complex fluctuation patterns. To enable wind power prediction models to process data more precisely, decomposition methods are widely used. By decomposing wind power sequences to obtain the corresponding sub-series models, complex fluctuation patterns can be extracted from the wind power sequences to improve prediction performance. However, after decomposing the different fluctuation representations of wind power, existing technologies usually directly input the decomposed sub-signals representing the different wind power fluctuations into the wind power prediction model for prediction. This method ignores the global interaction between the sub-signals when performing power prediction, resulting in poor wind power prediction results.

[0065] Therefore, in order to solve the problem that the method of fitting the linear or nonlinear relationship between historical meteorological data and historical power data does not take into account the complex fluctuation pattern of wind power and has inaccurate power prediction, the above steps S100 to S600 effectively retain the global interaction characteristics of the sub-signals through a three-stage collaborative mechanism of complete empirical mode decomposition, clustering and secondary decomposition: the first historical power data is subjected to complete empirical mode decomposition to obtain multiple intrinsic mode functions, and the intrinsic mode functions are divided into clusters with similar dynamic characteristics using a clustering method. While retaining the time-frequency correlation within the cluster, the potential coupling between different fluctuation modes is captured. Combined relationship; Based on the frequency characteristics of each cluster, each inherent mode function is decomposed twice. The secondary decomposition sub-signal retains the global interaction and local details of the first historical wind power in different frequency domains, and can effectively extract the complex fluctuation pattern in the wind power time series; The secondary decomposition sub-signal is used to perform power prediction to obtain the power prediction result at the future moment. The preset prediction model can extract the local fluctuation characteristics and global interaction of each inherent mode function by extracting the sequential causal relationship and reverse time information contained in each secondary decomposition sub-signal, so as to perform power prediction based on the local fluctuation characteristics and global interaction, and the prediction result is more accurate.

[0066] Example 2: Based on the previous example, this example introduces a specific implementation of a wind power prediction method based on secondary decomposition and reconstruction, in order to illustrate the technical solution of this method.

[0067] S100: Acquire first historical power data of a wind farm;

[0068] In an embodiment of the present application, the first historical power data may be obtained by collecting power from a power recording device of a target wind farm, wherein a time scale of the first historical power data is not less than a time scale of predicted wind power.

[0069] S200: performing complete empirical mode decomposition on the first historical power data to obtain a plurality of intrinsic mode functions of different frequencies corresponding to the first historical power data;

[0070] In an embodiment of the present application, the first historical power data is decomposed into a plurality of intrinsic mode functions that characterize different power fluctuation characteristics at different time scales and at different frequencies by complete empirical mode decomposition;

[0071] It should be noted that complete empirical mode decomposition is a signal processing technique used to decompose complex time series data into a series of simple intrinsic mode functions with different frequencies and a residual term; CEEMDAN is an extension of empirical mode decomposition (EMD), which improves the accuracy and reliability of the decomposition by introducing an ensemble averaging process and adaptive noise.

[0072] In an embodiment of the present application, when performing complete empirical mode decomposition on the first historical power data, multiple Gaussian white noises are first added to the first historical power data to construct multiple sub-historical power data sequences; further, EMD decomposition is performed on each sub-historical power data sequence to obtain an initial intrinsic mode function and a residual sequence; wherein the initial intrinsic mode function is obtained by averaging the intrinsic mode functions obtained by decomposing each sub-historical power data sequence, and the residual sequence is obtained by taking the difference between each sub-historical power data sequence and the initial intrinsic mode function.

[0073] After obtaining the residual sequence, multiple rounds of EMD decomposition are performed on the residual sequence. In each round of EMD decomposition, multiple Gaussian white noises are added to the residual sequence obtained by the previous round of decomposition to obtain multiple sub-residual sequences; EMD decomposition is performed on each sub-residual sequence to obtain the iterative intrinsic mode function of the current round and the residual sequence of the current round until the preset number of iterations is reached. The residual sequence obtained in the last round of EMD decomposition is used as the residual term, and the iterative intrinsic mode function and the initial intrinsic mode function obtained in each round of EMD decomposition are used as the intrinsic mode function.

[0074] It should be noted that by processing the first historical power data using complete empirical mode decomposition, it is possible to extract the fluctuation patterns of wind power from different frequency domains.

[0075] S300: Clustering each intrinsic mode function based on frequency characteristics to obtain multiple clusters;

[0076] It should be noted that the intrinsic mode function obtained by complete empirical mode decomposition can reflect the fluctuation form of wind power in different frequency domains; wherein, the fluctuation form is usually caused by different meteorological conditions. Therefore, among the intrinsic mode functions obtained by complete empirical mode decomposition, multiple similar intrinsic mode function clusters can reflect the fluctuation of wind power under a certain weather type; therefore, in the embodiment of the present application, in order to extract the intrinsic mode functions that belong to the same type of weather type under different weather types, each intrinsic mode function is clustered based on the frequency characteristics to obtain multiple cluster clusters.

[0077] Specifically, when clustering each intrinsic mode function based on frequency characteristics to obtain multiple clusters, the intrinsic mode functions can be clustered based on the similarity of the frequency characteristics of each intrinsic mode function to obtain at least two target clusters; wherein the target frequency domain intervals represented by each target cluster do not overlap.

[0078] Specifically, when clustering each intrinsic mode function based on frequency characteristics to obtain multiple clusters, the intrinsic mode functions can be clustered based on the similarity of the frequency characteristics of each intrinsic mode function to obtain a first target cluster, a second target cluster and a third target cluster; wherein the first target frequency domain within the first target cluster is greater than the second target frequency domain of the second target cluster, and the second target frequency domain of the second target cluster is greater than the third target frequency domain of the third target cluster.

[0079] Specifically, after obtaining the intrinsic modal function, we can first determine the influencing factors of each frequency factor contained in the frequency feature in characterizing the intrinsic modal function and determine the clustering weight of each frequency factor. We can further cluster the intrinsic modal function based on the clustering weight of the frequency factor and the corresponding numerical value to obtain the first target cluster cluster, the second target cluster cluster and the third target cluster cluster.

[0080] In an optional embodiment, a graph neural network method can be used to achieve clustering of intrinsic mode functions; specifically, nodes and edges of an undirected graph are constructed based on each intrinsic mode function, wherein each node is used to represent the time-frequency characteristics of each intrinsic mode function, and the time-frequency characteristics may include instantaneous frequency, energy distribution, and waveform similarity, etc. The characteristics of the edges of the undirected graph are defined as the initial similarity of each intrinsic mode function, and the edges of the undirected graph are determined based on the initial similarity of each intrinsic mode function to obtain a constructed undirected graph; wherein the initial similarity can be determined by a Gaussian kernel function, and further, the constructed undirected graph is input into the graph neural network, and the edge weights are learned through the attention mechanism based on the graph neural network to achieve clustering of the intrinsic mode functions.

[0081] In an optional embodiment, the spectral clustering (SC) method can be used to cluster the intrinsic mode functions. SC is a clustering algorithm based on graph theory that uses data similarity information to group data points into different categories. The core idea of spectral clustering is to transform the clustering problem into a graph partitioning problem, and to cluster data points by analyzing the graph spectrum (i.e., eigenvalues and eigenvectors).

[0082] In an embodiment of the present application, during spectral clustering, a similarity matrix is first constructed based on the time-frequency characteristics of the intrinsic mode functions, the morphological correlation of sequences corresponding to different intrinsic mode functions is measured by dynamic time warping, and an adaptive Gaussian kernel function is used to fuse frequency domain distance and time domain similarity to define edge weights between nodes; wherein, the time-frequency characteristics may include instantaneous frequency, energy distribution, and waveform similarity, etc.; further, the similarity matrix is converted into a normalized Laplace matrix, and the eigenvectors in the low-dimensional embedding space are obtained by eigenvalue decomposition, and the eigenvector space is clustered using the K-means algorithm, and finally the intrinsic mode functions are divided into three clusters: high frequency, medium frequency, and low frequency; this process effectively captures the potential coupling relationship between the intrinsic mode functions through graph structure modeling, so that the sub-signals within the same cluster maintain local frequency consistency and retain cross-modal interactions through global graph connections, providing a signal grouping basis with both frequency domain discrimination and time domain correlation for subsequent secondary decomposition.

[0083] It should be noted that the above step S300 performs complete empirical mode decomposition on the first historical power data to obtain multiple intrinsic mode functions, and uses a clustering method to divide the intrinsic mode functions into clusters with similar dynamic characteristics, while retaining the time-frequency correlation within the cluster, capturing the potential coupling relationship between different fluctuation modes.

[0084] S400: performing secondary decomposition on each intrinsic mode function based on the frequency characteristics of each cluster to obtain a secondary decomposition sub-signal corresponding to each cluster;

[0085] In an embodiment of the present application, each intrinsic mode function is subjected to a secondary decomposition based on the frequency characteristics of each cluster to obtain a secondary decomposition sub-signal corresponding to each cluster; wherein each secondary decomposition sub-signal is obtained by extracting a target element in a sub-time series whose actual influencing factor is greater than a preset factor, and the sub-time series is constructed by the intrinsic mode function to obtain a secondary decomposition sub-signal that retains the local characteristics and global characteristics of each cluster, so that when power prediction is subsequently performed through the secondary decomposition sub-signal, the local characteristics and global interactions of the decomposed sub-signals in the current frequency domain contained in the cluster are considered to perform power prediction.

[0086] In an optional embodiment, a singular spectrum analysis (SSA) method is used to perform a secondary decomposition of each intrinsic mode function. SSA is a method for processing nonlinear time series data. By performing operations such as decomposition and reconstruction on the trajectory matrix of the time series to be studied, different component sequences in the time series are extracted, thereby analyzing the time series.

[0087] In another optional embodiment, the mean decomposition method is used to perform secondary decomposition on each intrinsic mode function. By gradually separating the local extreme points and the mean curve of the signal, the product function (PF) components are generated, which is suitable for processing non-stationary wind power data.

[0088] In the embodiment of the present application, the specific steps for obtaining the secondary decomposition sub-signals corresponding to each clustering cluster include:

[0089] Convert each intrinsic mode function into a trajectory matrix with a sliding window of a preset window size;

[0090] Perform singular value decomposition on the trajectory matrix to obtain the singular values and orthogonal matrices of the trajectory matrix;

[0091] Sort and screen the singular values, and take the first N largest singular values as the target singular values;

[0092] Select target singular vectors based on the target singular values, and construct multiple sub-target sequences based on the target singular values, target singular vectors, and orthogonal matrices;

[0093] Perform diagonal averaging on each sub-target sequence to obtain the secondary decomposition sub-signal.

[0094] Specifically, the steps for performing secondary decomposition on each intrinsic mode function through the frequency characteristics of each clustering cluster by the SSA method include: converting each intrinsic mode function into a trajectory matrix with a sliding window of a preset window size L (1 < L < n), and each row of the trajectory matrix is a subsequence of the time series; performing singular value decomposition on the trajectory matrix to obtain the singular values and orthogonal matrices of the trajectory matrix; sorting and screening the singular values, and taking the first N largest singular values as the target singular values; selecting target singular vectors based on the target singular values, and constructing r subsequences, each subsequence is obtained by a linear combination of the column vectors in U and V; performing diagonal averaging on each sub-target sequence to obtain the final decomposition result; where diagonal averaging is to rearrange the elements in the subsequence in a diagonal manner and then take the average value.

[0095] It should be noted that singular spectrum analysis is used to perform secondary decomposition on the clustering cluster. By constructing a cross-modal trajectory matrix and extracting co-principal components, the phase synchronization characteristics and cross-scale fluctuation laws between sub-signals in the same frequency domain are explicitly captured; finally, a dual-channel fusion architecture is designed in the prediction model, which not only models the independent evolution modes of each subsequence through the local feature channel, but also dynamically learns the global interaction weights between different frequency domains with the help of the graph attention mechanism, thus achieving an organic unity of local feature fidelity and global information transmission at multiple levels; the above method significantly improves the utilization rate of the interaction characteristics between sub-signals through a progressive design of spectrum correlation preservation, cross-modal collaborative decomposition, and interaction-aware prediction compared with the traditional isolated processing strategy.

[0096] It should be noted that the secondary decomposition sub-signal corresponding to each cluster in the above step S400 not only contains the local fluctuation characteristics of the wind power in the frequency domain corresponding to the current cluster, but also contains the global interaction of the wind power in the frequency domain corresponding to the cluster, which can effectively extract the complex fluctuation pattern in the wind power time series, so as to improve the prediction performance of the power prediction model in the subsequent power prediction.

[0097] S500: Inputting each secondary decomposition sub-signal into a pre-built preset prediction model to perform power prediction, and obtaining a sub-power prediction result corresponding to each secondary decomposition sub-signal;

[0098] In the embodiment of the present application, the step of obtaining the sub-power prediction result corresponding to each secondary decomposition sub-signal includes:

[0099] Input each secondary decomposition sub-signal into a pre-trained preset prediction model, and extract the causal dependency and reverse time information of each secondary decomposition sub-signal based on the preset prediction model;

[0100] Reversing each reverse time information into reverse time information;

[0101] The sub-power prediction results are obtained based on the causal relationship and flip time information.

[0102] In an optional embodiment, the preset prediction model may include a time feature extraction module, a time processing module and a power prediction module; wherein the time feature extraction module may include a first feature extraction unit and a second feature extraction unit connected in parallel, the first feature extraction unit is used to extract the causal dependency of each secondary decomposition sub-signal, and the second feature extraction unit is used to extract the reverse time information of the secondary decomposition sub-signal; the time processing module is used to flip the reverse time information to obtain flipped time information to keep consistent with the forward representation in the time dimension; the power prediction module is used to map the causal relationship and flipped time information into sub-power prediction results.

[0103] Exemplarily, the first feature extraction unit may adopt a causal convolutional network or a Transformer encoder to model the forward temporal evolution law through a masked attention mechanism.

[0104] Exemplarily, the second feature extraction unit may use the reverse layer of a bidirectional long short-term memory network or the reverse causal convolution layer of a temporal convolutional network to capture the reverse dependency features of the sequence.

[0105] Exemplarily, the time processing module is used to flip the reverse time information to obtain flipped time information to keep consistent with the forward representation in the time dimension; exemplarily, the symmetrical alignment of the timing dimension can be achieved through a tensor flip operation layer or a reverse sequence reorganization layer.

[0106] Exemplarily, the power prediction module is used to map causal relationships and flipping time information into sub-power prediction results. Specifically, the power prediction module can integrate the gated attention fusion layer and the fully connected output layer, dynamically weightedly fuse bidirectional temporal features, and generate the final prediction value through nonlinear mapping.

[0107] In another optional embodiment, the preset prediction model may be a TSMamaba model; the TSMamaba model is a sequence modeling architecture that uses a selective state space model to optimize the efficiency and performance of processing long sequence data; the Mamba model introduces a selective mechanism to maintain the complexity of the selective state space model while maintaining performance similar to that of the Transformer.

[0108] In an embodiment of the present application, the preset prediction model is the TSMamba model, which includes a cascaded bidirectional encoder module, a timing alignment module and a power output module. The bidirectional encoder module includes a forward encoder and a backward encoder; the bidirectional encoder module includes a forward encoder and a reverse encoder connected in parallel; the forward encoder is used to extract the causal dependency of each secondary decomposition sub-signal; the reverse time information is used to extract the reverse time information of each secondary decomposition sub-signal; the timing alignment module is used to flip the reverse time information to obtain the flip time information; the power output module is used to map the causal relationship and flip time information into sub-power prediction results.

[0109] In an embodiment of the present application, the TSMamba model is used to implement wind power prediction based on quadratic decomposition sub-signals; the TSMamba prediction framework combines the Transformer algorithm and the Mamba algorithm, and adopts a bidirectional encoder design. The forward encoder is used to capture causal dependencies, and the backward encoder extracts reverse time information to maximize the prediction capability and processing efficiency to obtain sub-power prediction results.

[0110] It should be noted that in the above step S500, since each secondary decomposition sub-signal contains the local fluctuation characteristics of the wind power in the frequency domain corresponding to the current cluster, it also contains the global interaction of the wind power in the frequency domain corresponding to the cluster. Therefore, the preset prediction model can extract the local fluctuation characteristics and global interaction of each inherent mode function by extracting the sequential causal relationship and reverse time information contained in each secondary decomposition sub-signal, fit the secondary decomposition sub-signal to capture the local and global characteristics in the specific frequency domain, and further perform power prediction based on the local fluctuation characteristics and global interaction to obtain sub-power prediction results in different frequency domains.

[0111] S600: Fusing the sub-power prediction results to obtain a wind power prediction result.

[0112] In the embodiment of the present application, the specific steps of fusing the sub-power prediction results to obtain the wind power prediction result include:

[0113] Obtain the start and end times of each sub-power prediction result;

[0114] The start time and the end time are input into the error prediction model to obtain the prediction error of each sub-power prediction result; wherein the error prediction model is trained based on the time information and time series dependency corresponding to the historical prediction errors of the preset prediction model. During the error prediction model training process, the correspondence between the moment changes of the historical prediction errors within the time information and the time series dependency is learned;

[0115] Based on the prediction error and the prediction results of each sub-power, the ridge regression method is used to obtain the wind power prediction results.

[0116] In the embodiment of the present application, the specific steps of training the error prediction model include:

[0117] Obtaining a historical sub-prediction result of a preset prediction model and second historical power data corresponding to a time series of the historical sub-prediction result; wherein the time of the second historical power is earlier than the first historical power data;

[0118] Obtain a prediction error sequence with a timestamp based on the historical sub-prediction results and the second historical power data;

[0119] The prediction error sequence is input into a pre-built preset correction model for model training. During the model training process, the correspondence between the moment changes of the prediction error sequence in the time interval corresponding to the timestamp and the timing dependency is learned until the model converges to obtain the error prediction model.

[0120] In an optional embodiment, the error prediction model may be at least one of a recurrent neural network, a long-short-time neural network, and a temporal convolutional network; the error prediction model is used to learn the correspondence between the moment changes of the prediction error sequence in the time interval corresponding to the timestamp and the temporal dependency.

[0121] In an embodiment of the present application, the error prediction model is a temporal convolutional network, which models and predicts the prediction error of the preset prediction model through TCN, and can capture the temporal change information in the error sequence composed of the historical errors of the preset prediction model. Finally, the wind power prediction result is obtained by combining the prediction results of each secondary decomposition sub-signal and the corresponding prediction error.

[0122] It should be noted that the above-mentioned step S600 uses the secondary decomposition sub-signal to perform power prediction to obtain the power prediction result at the future moment. The preset prediction model can extract the local fluctuation characteristics and global interactions of each inherent mode function by extracting the sequential causal relationship and reverse time information contained in each secondary decomposition sub-signal, so as to perform power prediction based on the local fluctuation characteristics and global interaction, and the prediction result is more accurate.

[0123] Example 3: This example provides a wind power prediction system based on secondary decomposition and reconstruction, including:

[0124] A data acquisition module, configured to acquire first historical power data of a wind farm;

[0125] a first signal decomposition module, configured to perform complete empirical mode decomposition on the first historical power data to obtain a plurality of intrinsic mode functions of different frequencies corresponding to the first historical power data;

[0126] A clustering module is used to cluster each intrinsic mode function based on frequency characteristics to obtain multiple clusters;

[0127] The second signal decomposition module is used to perform secondary decomposition on each intrinsic mode function based on the frequency characteristics of each cluster to obtain a secondary decomposition sub-signal corresponding to each cluster;

[0128] The sub-power prediction module is used to input each secondary decomposition sub-signal into a pre-built preset prediction model to perform power prediction and obtain the sub-power prediction result corresponding to each secondary decomposition sub-signal;

[0129] The power fusion module is used to fuse the sub-power prediction results to obtain the wind power prediction results.

[0130] It should be noted that the technical solution of the wind power prediction system based on secondary decomposition and reconstruction and the technical solution of the wind power prediction method based on secondary decomposition and reconstruction mentioned above belong to the same concept. For the details not described in detail in the technical solution of the wind power prediction system based on secondary decomposition and reconstruction in this embodiment, please refer to the description of the technical solution of the wind power prediction method based on secondary decomposition and reconstruction mentioned above.

[0131] The above-mentioned unit modules can be embedded in or independent of the processor in the electronic device in the form of hardware, or can be stored in the memory of the electronic device in the form of software, so that the processor can call and execute the corresponding operations of the above-mentioned modules.

[0132] This embodiment also provides an electronic device, which includes a processor, memory, a communication interface, a display screen, and an input device connected via a system bus. The processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The communication interface of the electronic device is used to communicate with an external terminal in a wired or wireless manner. The wireless manner can be achieved through WIFI, an operator network, NFC (near field communication), or other technologies. When the computer program is executed by the processor, a wind power prediction method based on secondary decomposition and reconstruction is implemented. The display screen of the electronic device can be a liquid crystal display screen or an electronic ink display screen. The input device of the electronic device can be a touch layer covering the display screen, or a key, trackball, or touchpad provided on the housing of the electronic device, or an external keyboard, touchpad, or mouse.

[0133] This embodiment further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method proposed in the above embodiment is implemented.

[0134] The storage medium proposed in this embodiment and the method proposed in the above embodiment belong to the same inventive concept. For technical details not fully described in this embodiment, please refer to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0135] Through the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented with the help of software and necessary general-purpose hardware, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for enabling an electronic device (which can be a personal computer, server, or network device, etc.) to execute the method of the embodiment of the present invention.

[0136] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

[0137] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiment of the present application can be implemented in various computer languages.

[0138] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0139] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0140] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0141] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0142] Obviously, those skilled in the art may make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is intended to include these modifications and variations.

Claims

1. A wind power prediction method based on secondary decomposition and reconstruction is characterized by: include: Obtaining first historical power data of a wind farm; Performing complete empirical mode decomposition on the first historical power data to obtain a plurality of intrinsic mode functions of different frequencies corresponding to the first historical power data; Clustering the intrinsic mode functions based on frequency characteristics to obtain multiple clusters; Performing a secondary decomposition on each of the intrinsic mode functions based on the frequency characteristics of each of the clusters to obtain a secondary decomposition sub-signal corresponding to each of the clusters; Inputting each of the secondary decomposition sub-signals into a pre-built preset prediction model to perform power prediction, and obtaining a sub-power prediction result corresponding to each of the secondary decomposition sub-signals; The sub-power prediction results are integrated to obtain a wind power prediction result.

2. The wind power prediction method based on secondary decomposition and reconstruction according to claim 1, characterized in that: The clustering of the intrinsic mode functions based on the frequency characteristics to obtain a plurality of clusters includes: Extracting time-frequency features of each intrinsic mode function to obtain time-frequency features corresponding to each intrinsic mode function; Clustering the intrinsic mode functions based on the similarity of the frequency characteristics of each of the intrinsic mode functions to obtain a first target cluster, a second target cluster, and a third target cluster; The first target frequency domain in the first target cluster is larger than the second target frequency domain in the second target cluster, and the second target frequency domain in the second target cluster is larger than the third target frequency domain in the third target cluster.

3. The wind power prediction method based on secondary decomposition and reconstruction according to claim 2, characterized in that: The obtaining of the secondary decomposition sub-signals corresponding to the clusters includes: Converting each of the intrinsic mode functions into a trajectory matrix using a sliding window of a preset window size; Performing singular value decomposition on the trajectory matrix to obtain singular values and an orthogonal matrix of the trajectory matrix; Sorting and screening the singular values to obtain target singular values; Selecting a target singular vector based on the target singular value, and constructing a plurality of sub-target sequences based on the target singular value, the target singular vector and the orthogonal matrix; Diagonal averaging is performed on each of the sub-target sequences to obtain the secondary decomposition sub-signal.

4. The wind power prediction method based on secondary decomposition and reconstruction according to claim 3, characterized in that: Obtaining the sub-power prediction results corresponding to each of the secondary decomposition sub-signals includes: Inputting each of the secondary decomposition sub-signals into a pre-trained preset prediction model, and extracting the causal dependency and reverse time information of each of the secondary decomposition sub-signals based on the preset prediction model; Reversing each of the reverse time information into reverse time information; The sub-power prediction result is obtained based on the causal relationship and the flip time information.

5. The wind power prediction method based on secondary decomposition and reconstruction according to claim 4, characterized in that: The preset prediction model includes a cascaded bidirectional encoder module, a timing alignment module and a power output module; The bidirectional encoder module includes a forward encoder and a reverse encoder connected in parallel, the forward encoder is used to extract the causal dependency of each of the secondary decomposition sub-signals, and the reverse encoder is used to extract reverse time information of each of the secondary decomposition sub-signals; The timing alignment module is used to flip the reverse time information to obtain the flip time information; The power output module is used to map the causal relationship and the reversal time information into the sub-power prediction result.

6. The wind power prediction method based on secondary decomposition and reconstruction according to claim 5, characterized in that: The fusing of the sub-power prediction results to obtain a wind power prediction result includes: Obtaining the start time and end time of each sub-power prediction result; Inputting the start time and the end time into an error prediction model to obtain a prediction error of each sub-power prediction result; wherein the error prediction model is trained based on the time information and timing dependency corresponding to the historical prediction errors of the preset prediction model, and during the error prediction model training process, the corresponding relationship between the moment changes of the historical prediction errors within the time information and the timing dependency is learned; Based on the prediction error and each of the sub-power prediction results, a ridge regression method is used to obtain the wind power prediction result.

7. The wind power prediction method based on secondary decomposition and reconstruction according to claim 6, characterized in that: The training of the error prediction model includes: Obtaining a historical sub-prediction result of the preset prediction model and second historical power data corresponding to the time series of the historical sub-prediction result; wherein the time of the second historical power is earlier than that of the first historical power data; Obtaining a prediction error sequence with a timestamp based on the historical sub-prediction results and the second historical power data; The prediction error sequence is input into a pre-built preset correction model for model training. During the model training process, the correspondence between the moment changes of the prediction error sequence in the time interval corresponding to the timestamp and the timing dependency is learned until the model converges to obtain the error prediction model.

8. A wind power prediction system based on secondary decomposition and reconstruction, applying the wind power prediction method based on secondary decomposition and reconstruction according to any one of claims 1 to 7, characterized in that: include: A data acquisition module, configured to acquire first historical power data of a wind farm; a first signal decomposition module, configured to perform complete empirical mode decomposition on the first historical power data to obtain a plurality of intrinsic mode functions of different frequencies corresponding to the first historical power data; A clustering module, configured to cluster the intrinsic mode functions based on frequency characteristics to obtain a plurality of clusters; A second signal decomposition module is used to perform secondary decomposition on each of the intrinsic mode functions based on the frequency characteristics of each of the clusters to obtain secondary decomposition sub-signals corresponding to each of the clusters; A sub-power prediction module is used to input each of the secondary decomposition sub-signals into a pre-built preset prediction model to perform power prediction, and obtain a sub-power prediction result corresponding to each of the secondary decomposition sub-signals; The power fusion module is used to fuse the sub-power prediction results to obtain a wind power prediction result.

9. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store computer-executable instructions, and the processor implements the steps of the wind power prediction method based on secondary decomposition and reconstruction according to any one of claims 1 to 7 when executing the computer-executable instructions.

10. A computer-readable storage medium having computer-executable instructions stored thereon, characterized in that: When the computer executable instructions are executed by a processor, the steps of the wind power prediction method based on secondary decomposition and reconstruction according to any one of claims 1 to 7 are implemented.

Images