Power prediction model construction method and device based on mutual information and signal decomposition, computer equipment and storage medium

Abstract

The invention relates to the technical field of clean energy, and discloses a power prediction model construction method based on mutual information and signal decomposition, and the method comprises the steps: obtaining historical meteorological data and historical power data of a wind power station; performing mutual information calculation on the historical meteorological data and the historical power data to obtain influence weights of the historical meteorological data relative to the historical power data; performing complete empirical mode decomposition on the historical power data to obtain a signal decomposition result; performing feature screening and feature reconstruction based on a signal decomposition result to obtain a reconstructed intrinsic mode function and a residual term; inputting the historical meteorological data, the historical power data and the reconstructed intrinsic mode function into a pre-constructed preset model, carrying out model training under the constraint of the influence weight until the model converges, and obtaining a power prediction model; according to the method, the power prediction model performs power prediction by considering the power fluctuation condition under the extreme weather condition, so that the accuracy and robustness of the power prediction model are improved.

Description

Technical Field

[0001] The present invention relates to the field of clean energy technology, and in particular to a method, device, computer equipment and storage medium for constructing a power prediction model based on mutual information and signal decomposition. Background Art

[0002] As an important energy source, wind power is highly sensitive to changes in weather conditions and exhibits strong volatility and randomness.

[0003] Traditional power prediction models rely on historical data and conventional numerical weather forecasts. These models can maintain a certain degree of prediction accuracy under normal weather conditions. However, in extreme weather conditions, the volatility and randomness of power load are amplified. Due to insufficient analysis of extreme weather, the prediction accuracy of power prediction models that rely on historical data and conventional numerical weather forecasts often drops significantly, and they cannot meet the high standards required by the power grid. Summary of the Invention

[0004] In view of this, the present invention provides a method for constructing a power prediction model based on mutual information and signal decomposition to solve the problem that power prediction models that rely on historical data and conventional numerical weather forecasts are inaccurate due to insufficient analysis of extreme weather.

[0005] In the first aspect, the present invention provides a method for constructing a power prediction model based on mutual information and signal decomposition, characterized in that the method for constructing a power prediction model based on mutual information and signal decomposition includes: obtaining historical meteorological data and historical power data of a wind farm station; wherein the historical meteorological data includes time series data corresponding to multiple types of meteorological factors; performing mutual information calculation on the historical meteorological data and historical power data to obtain the influence weight of each historical meteorological data relative to the historical power data; performing complete empirical mode decomposition on the historical power data to obtain a signal decomposition result; wherein the signal decomposition result includes a plurality of intrinsic mode functions with different frequencies and residuals corresponding to the signal decomposition result Item; each of the intrinsic mode functions is used to characterize the historical fluctuations of the historical power data at different time scales, and the residual term is used to characterize the overall trend of the historical power data; feature screening and feature reconstruction are performed based on the signal decomposition result to obtain reconstructed intrinsic mode functions and residual terms; the historical meteorological data, the historical power data, the reconstructed intrinsic mode functions and the residual terms are input into a pre-built preset model, and the model is trained under the constraint of the influence weight until the model converges to obtain a power prediction model; in the process of model training, the correspondence between the historical meteorological data and the reconstructed intrinsic mode functions, the residual terms and the historical power data is learned.

[0006] As an exemplary embodiment, the mutual information calculation of the historical meteorological data and the historical power data to obtain the influence weight of each historical meteorological data relative to the historical power data includes: performing mutual information calculation on the historical meteorological data and the historical power data to obtain the joint probability distribution and marginal probability distribution of each meteorological factor contained in the historical meteorological data relative to the historical power data; calculating the information entropy of each meteorological factor relative to the historical power data based on the joint probability distribution and the marginal probability distribution; determining the influence weight of each historical meteorological data relative to the historical power data based on the information entropy; wherein the influence weight is proportional to the information entropy.

[0007] As an exemplary embodiment, before inputting the historical meteorological data, historical power data, the reconstructed intrinsic mode function and the residual term into a pre-built preset model, the method for constructing a power prediction model based on mutual information and signal decomposition also includes: screening the historical meteorological data based on the information entropy to obtain target meteorological data; wherein the information entropy of each meteorological factor contained in each target meteorological data is greater than a first preset value; and constructing a training data set based on the target meteorological data, the historical power data, the reconstructed intrinsic mode function and the residual term.

[0008] As an exemplary embodiment, the historical power data is subjected to complete empirical mode decomposition to obtain a signal decomposition result, including: adding multiple Gaussian white noises to the historical power data to construct multiple sub-historical power data sequences; performing EMD decomposition on each of the sub-historical power data sequences 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 of the sub-historical power data sequences, and the residual sequence is obtained by taking the difference between each of the sub-historical power data sequences and the initial intrinsic mode function; performing EMD on the residual sequence Perform multiple rounds of EMD decomposition, and in each round of EMD decomposition, add multiple Gaussian white noises to the residual sequence obtained in the previous round of decomposition to obtain multiple sub-residual sequences; perform EMD decomposition on each of the sub-residual sequences to obtain the iterative intrinsic mode function of the current round and the residual sequence of the current round, until a preset number of iterations is reached, and use the residual sequence obtained in the last round of EMD decomposition as the residual term, and use the iterative intrinsic mode function and the initial intrinsic mode function obtained in each round of EMD decomposition as the intrinsic mode function to obtain the signal decomposition result.

[0009] As an exemplary embodiment, the preset model includes a first feature extraction module, a second feature extraction module and a power output module, and the historical meteorological data, the historical power data, the reconstructed intrinsic mode function and the residual term are input into the pre-built preset model, and the model training is performed under the constraint of the influence weight; including: inputting the historical meteorological data, the reconstructed intrinsic mode function and the residual term into the preset model for model training, continuously adjusting the parameters of the first feature extraction module so that the first feature extraction module can learn the correspondence between the historical meteorological data and the reconstructed intrinsic mode function and the residual term under the influence weight; continuously adjusting the parameters of the second feature extraction module so that the second feature extraction module can learn the correspondence between the local time series change features and the global time series change features extracted from the reconstructed intrinsic mode function and the residual term; continuously adjusting the parameters of the power output module so that the power output module learns the correspondence between the local time series change features and the global time series change features and the historical power data.

[0010] As an exemplary embodiment, the preset power prediction model is a CNTTansNet model.

[0011] As an exemplary embodiment, the first feature extraction module includes at least one of CNN and RNN, and the second feature extraction module includes at least one of Informer and Transformer.

[0012] In the second aspect, the present invention provides a power prediction model construction device based on mutual information and signal decomposition, and the power prediction model construction device based on mutual information and signal decomposition includes: an acquisition module for acquiring historical meteorological data and historical power data of a wind farm station; wherein the historical meteorological data includes time series data corresponding to multiple types of meteorological factors; a calculation module for performing mutual information calculation on the historical meteorological data and historical power data to obtain the influence weight of each of the historical meteorological data relative to the historical power data; a signal decomposition module for performing complete empirical mode decomposition on the historical power data to obtain a signal decomposition result; wherein the signal decomposition result includes a plurality of intrinsic mode functions with different frequencies and a pair of signal decomposition results. corresponding residual terms; each of the intrinsic mode functions is used to characterize the historical fluctuations of the historical power data at different time scales, and the residual terms are used to characterize the overall trend of the historical power data; a feature reconstruction module is used to perform feature screening and feature reconstruction based on the signal decomposition result to obtain reconstructed intrinsic mode functions and residual terms; a model training module is used to input the historical meteorological data, historical power data and the reconstructed intrinsic mode functions into a pre-built preset model, and perform model training under the constraints of the influence weights until the model converges to obtain a power prediction model; during the model training process, the correspondence between the historical meteorological data and the reconstructed intrinsic mode functions, the residual terms and the historical power data is learned.

[0013] In a third aspect, the present invention provides a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the method of the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0014] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the method of the first aspect or any corresponding embodiment thereof.

[0015] The power prediction model construction method based on mutual information and signal decomposition of the present invention comprises: obtaining historical meteorological data and historical power data of a wind farm station; wherein, the historical meteorological data comprises time series data corresponding to multiple types of meteorological factors; performing mutual information calculation on the historical meteorological data and historical power data to obtain the influence weight of each of the historical meteorological data relative to the historical power data; performing complete empirical mode decomposition on the historical power data to obtain a signal decomposition result; wherein, the signal decomposition result comprises a plurality of intrinsic mode functions with different frequencies and residual terms corresponding to the signal decomposition result; each of the intrinsic mode functions is used to characterize the historical fluctuation of the historical power data at different time scales, and the residual term is used to characterize the overall trend of the historical power data; performing feature screening and feature reconstruction based on the signal decomposition result to obtain a reconstructed intrinsic mode function and a residual term; inputting the historical meteorological data, historical power data and the reconstructed intrinsic mode function into a pre-constructed preset model, and The model is trained under the constraint of the response weight until the model converges to obtain a power prediction model; in the process of model training, the correspondence between the historical meteorological data and the reconstructed intrinsic mode function and the historical power data is learned; in the feature extraction stage, the above method can determine the influence of each meteorological factor contained in the historical meteorological data on the power prediction through mutual information calculation and empirical mode decomposition, and the obtained reconstructed intrinsic mode function can represent the power fluctuation on different time scales, which includes the power fluctuation state under extreme weather conditions; and the residual term can reflect the characteristic performance of the historical power data on a long time scale; the power prediction model obtained by training the correspondence between the historical meteorological data and the reconstructed intrinsic mode function and the historical power data improves the efficiency of data analysis and feature selection, especially for processing complex, nonlinear and non-stationary time series data, so that the power prediction model can consider the power fluctuation state under extreme weather conditions for power prediction, thereby improving the accuracy and robustness of the power prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] 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.

[0017] Figure 1 1 is a flow chart of a method for constructing a power prediction model based on mutual information and signal decomposition according to an embodiment of the present invention;

[0018] Figure 2 is a structural block diagram of a device for constructing a power prediction model based on mutual information and signal decomposition according to an embodiment of the present invention;

[0019] Figure 3 Schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0020] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of the present invention.

[0021] According to an embodiment of the present invention, an embodiment of a method for constructing a power prediction model based on mutual information and signal decomposition is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0022] Example 1: This embodiment provides a method for constructing a power prediction model based on mutual information and signal decomposition. Figure 1 FIG. 1 is a flow chart of a method for constructing a power prediction model based on mutual information and signal decomposition according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps:

[0023] Step S101: Acquire historical meteorological data and historical power data of a wind farm.

[0024] In this embodiment, historical meteorological data can be collected by meteorological collection equipment; illustratively, historical meteorological data includes time series data corresponding to multiple types of meteorological factors; specifically, historical meteorological data may include time series data corresponding to temperature, humidity, wind speed, and pressure.

[0025] In this embodiment, the historical power data can be obtained by acquiring the historical power recording device of the target wind farm.

[0026] Exemplarily, after obtaining historical meteorological data and historical power data, the historical meteorological data and historical power data are preprocessed; specifically, during the preprocessing process, missing values and outliers in the data are processed to ensure the accuracy and completeness of the data, and the data are normalized so that different variables can be compared on the same scale.

[0027] Step S102 : performing mutual information calculation on the historical meteorological data and the historical power data to obtain the influence weight of each historical meteorological data relative to the historical power data.

[0028] In related technologies, power prediction of wind farms is usually achieved by fitting historical meteorological data and historical power data collected by wind farms; when fitting the correspondence between historical meteorological data and historical power data, historical meteorological data is usually input into a wind power prediction model, and the parameters of the wind power prediction model are continuously adjusted so that the model can output a power value whose error with the historical power data meets the preset error, and further numerical weather forecast data for future time periods is input into the power prediction model to achieve power prediction.

[0029] However, on the one hand, historical meteorological data usually include multiple types of meteorological factors; under normal meteorological conditions, each type of meteorological factor contributes differently to the output power when predicting wind power, and some meteorological factors usually have an extreme impact on wind power under extreme meteorological conditions, for example; therefore, the power prediction model obtained by training without considering the contribution of various meteorological factors to wind power when predicting wind power has the problem of inaccuracy.

[0030] On the other hand, the correspondence between meteorological data and power data of wind farms under extreme weather conditions is different from that under normal weather conditions. Due to the randomness and volatility of wind resources, the wind power prediction method in related technologies can still maintain a certain prediction accuracy under normal weather conditions, but its prediction accuracy is greatly reduced under extreme weather conditions.

[0031] Moreover, affected by the randomness and intermittency of wind power generation, historical meteorological data and historical power data are usually reflected as complex, nonlinear and non-stationary time series data. Failure to consider the fluctuation characteristics of historical meteorological data and historical power data will also lead to inaccurate power forecasts.

[0032] Therefore, in this embodiment, the influence weight of each historical meteorological data relative to the historical power data is obtained through mutual information calculation, and the contribution of each meteorological factor contained in the historical meteorological data to the power forecast can be determined, so as to fully consider the complexity, nonlinearity and non-stationarity between the historical meteorological data and the power data.

[0033] In one embodiment, mutual information calculation is performed on the historical meteorological data and the historical power data to obtain the information entropy of the sub-historical meteorological data of each meteorological factor relative to the historical power data, and the influence weight of each historical meteorological data relative to the historical power data is further determined based on the information entropy.

[0034] In one embodiment, after obtaining the information entropy, feature screening is performed on the historical meteorological data based on the information entropy, meteorological data with a strong correlation with the historical power data is selected, and meteorological data with a weak correlation with the historical power data is eliminated.

[0035] Step S103, performing complete empirical mode decomposition on the historical power data to obtain a signal decomposition result; wherein the signal decomposition result includes multiple intrinsic mode functions with different frequencies and residual terms corresponding to the signal decomposition results; each of the intrinsic mode functions is used to characterize the historical fluctuations of the historical power data at different time scales, and the residual term is used to characterize the overall trend of the historical power data.

[0036] In this embodiment, the historical power data includes power data under normal weather conditions and power data under abnormal weather conditions. In order to extract the features corresponding to the two weather types contained in the historical power data, and further consider the power fluctuation features under the two weather types when making wind power forecasts, it is necessary to further extract features from the historical power data.

[0037] In this embodiment, feature extraction of historical power data is achieved by performing complete empirical mode decomposition on the historical power data; specifically, complete empirical mode decomposition is performed on the historical power data to obtain a signal decomposition result; wherein the signal decomposition result includes a plurality of intrinsic mode functions with different frequencies and residual terms corresponding to the signal decomposition results; each of the intrinsic mode functions is used to characterize the historical fluctuations of the historical power data at different time scales, and the residual term is used to characterize the overall trend of the historical power data.

[0038] In this embodiment, the historical power data is subjected to complete empirical mode decomposition to obtain a signal decomposition result. The inherent mode function contained in the signal decomposition result can reflect the fluctuation characteristics of the power data under different weather types; on the other hand, the residual term can reflect the characteristic performance of the historical power data on a long time scale.

[0039] Step S104 : performing feature screening and feature reconstruction based on the signal decomposition result to obtain a reconstructed intrinsic mode function and a residual term.

[0040] After obtaining the signal decomposition result, the signal decomposition result includes multiple intrinsic mode functions with different frequencies and residual terms corresponding to the signal decomposition result; wherein the intrinsic mode function may include sub-intrinsic mode functions under multiple types of frequencies; specifically, the intrinsic mode function may include multiple high-frequency sub-intrinsic mode functions reflecting short-term power fluctuations, multiple medium-frequency sub-intrinsic mode functions reflecting medium-term power fluctuations, and multiple low-frequency sub-intrinsic mode functions reflecting long-term power fluctuations; after obtaining the signal decomposition result, feature screening and reconstruction are performed based on the signal decomposition result to obtain a reconstructed intrinsic mode function, and a model is subsequently trained based on the reconstructed intrinsic mode function, so that when performing model training, multiple types of fluctuations of power data at multiple time scales can be considered for power prediction.

[0041] For example, when performing feature screening based on the signal decomposition results, multiple typical sub-inherent mode functions that can represent different meteorological conditions are selected in conjunction with the fluctuation mode of wind power; specifically, in order to enable the model to focus on the power performance under extreme weather conditions during subsequent training, when screening the sub-inherent mode functions under multiple frequency categories obtained by decomposition, on the one hand, multiple first low-frequency sub-inherent mode functions that can represent the power fluctuation situation under extreme weather conditions are selected; since the fluctuation situation of wind power under normal weather conditions is relatively stable, therefore, some of the power fluctuation situations under normal weather conditions are eliminated. On the other hand, the multiple second low-frequency sub-inherent modal functions reflecting the medium-term power fluctuations and the multiple low-frequency sub-inherent modal functions reflecting the long-term power fluctuations can characterize the power performance of the wind farm under normal weather and extreme weather conditions under the current geographical conditions and environmental conditions. When performing feature screening, the intermediate-frequency sub-inherent modal functions and the low-frequency sub-inherent modal functions are retained, so that the intrinsic modal functions finally screened can include both the short-term fluctuation characteristics of power under extreme weather conditions and the fluctuation characteristics of power over a long time scale.

[0042] Exemplarily, after obtaining multiple sub-intrinsic mode functions, each sub-intrinsic mode function is reconstructed to obtain a reconstructed intrinsic mode function that can be recognized by the model; in the signal decomposition and reconstruction process, the intrinsic mode functions (IMFs) and residual terms obtained by modal decomposition of the original wind power sequence are linearly superimposed to achieve signal reconstruction; specifically, each intrinsic mode function component characterizes the local fluctuation characteristics of different frequency domains, while the residual term retains the long-term trend information of the sequence; during reconstruction, all intrinsic mode function components and residual terms are inversely superimposed, and a weight coefficient is introduced to selectively attenuate the intrinsic mode function dominated by high-frequency noise to suppress overfitting interference in the reconstructed signal.

[0043] Exemplarily, after obtaining multiple sub-intrinsic mode functions, the multiple sub-intrinsic mode functions are clustered to obtain cluster decomposition result pairs; further, based on the clustering analysis results, the intrinsic mode functions in the same frequency domain are phase aligned and energy normalized to ensure the waveform consistency of cross-scale components when superimposed in the time domain. The reconstructed intrinsic mode function finally obtained not only completely retains the dynamic characteristics of the original sequence, but also optimizes the input quality of the subsequent prediction model through frequency domain feature enhancement.

[0044] The above method of this embodiment performs feature screening and feature reconstruction based on the signal decomposition results. On the one hand, the obtained reconstructed intrinsic mode function can represent the power fluctuations on different time scales, which includes the power fluctuation state under extreme weather conditions; on the other hand, the residual term can reflect the characteristic performance of historical power data on a long time scale.

[0045] Step S105: input the historical meteorological data, the historical power data, the reconstructed intrinsic mode function and the residual term into a pre-built preset model, and perform model training under the constraint of the influence weight until the model converges to obtain a power prediction model; during the model training process, learn the correspondence between the historical meteorological data and the reconstructed intrinsic mode function, the residual term and the historical power data.

[0046] In this embodiment, after obtaining the reconstructed intrinsic modal function, the historical meteorological data, historical power data and the reconstructed intrinsic modal function are input into a pre-built preset model, and model training is performed under the constraint of the influence weight. During the model training process, the correspondence between the historical meteorological data and the reconstructed intrinsic modal function, the residual term and the historical power data is learned.

[0047] The above-mentioned model training method performs model training based on the influence weights obtained by mutual information calculation, and can determine the contribution of various meteorological factors contained in historical meteorological data to power prediction, so as to fully consider the complexity, nonlinearity and non-stationarity between historical meteorological data and power data; and, considers reconstructing the intrinsic mode function for model training, and the reconstructed intrinsic mode function can represent power fluctuations on different time scales, which includes the power fluctuation state under extreme weather conditions; and, the residual term can reflect the characteristic performance of historical power data on a long time scale; the power prediction model obtained by training considering the correspondence between historical meteorological data and the reconstructed intrinsic mode function and historical power data improves the efficiency of data analysis and feature selection, especially for processing complex, nonlinear and non-stationary time series data; and, can enable the power prediction model to consider the power fluctuation state under extreme weather conditions for power prediction, thereby improving the accuracy and robustness of the power prediction model.

[0048] The power prediction model construction method based on mutual information and signal decomposition of this embodiment obtains historical meteorological data and historical power data of the wind farm station; wherein, the historical meteorological data includes time series data corresponding to multiple types of meteorological factors; mutual information calculation is performed on the historical meteorological data and historical power data to obtain the influence weight of each historical meteorological data relative to the historical power data; complete empirical mode decomposition is performed on the historical power data to obtain a signal decomposition result; wherein, the signal decomposition result includes multiple intrinsic mode functions with different frequencies and residual terms corresponding to the signal decomposition result; each intrinsic mode function is used to characterize the historical fluctuation of the historical power data at different time scales, and the residual term is used to characterize the overall trend of the historical power data; feature screening and feature reconstruction are performed based on the signal decomposition result to obtain reconstructed intrinsic mode function and residual term; the historical meteorological data, historical power data and the reconstructed intrinsic mode function are input into a pre-constructed preset model, and the influence weight is calculated based on the influence weight. The model is trained under the constraint of until the model converges to obtain a power prediction model; in the process of model training, the correspondence between the historical meteorological data and the reconstructed intrinsic mode function, the residual term and the historical power data is learned; in the feature extraction stage, the above method can determine the influence of each meteorological factor contained in the historical meteorological data on the power prediction through mutual information calculation and empirical mode decomposition, and the obtained reconstructed intrinsic mode function can represent the power fluctuation on different time scales, which includes the power fluctuation state under extreme weather conditions; and the residual term can reflect the characteristic performance of historical power data on a long time scale; the power prediction model obtained by training the correspondence between the historical meteorological data and the reconstructed intrinsic mode function and the historical power data improves the efficiency of data analysis and feature selection, especially for processing complex, nonlinear and non-stationary time series data, and can enable the power prediction model to consider the power fluctuation state under extreme weather conditions for power prediction, thereby improving the accuracy and robustness of the power prediction model.

[0049] As an exemplary embodiment, the mutual information calculation of the historical meteorological data and the historical power data to obtain the influence weight of each historical meteorological data relative to the historical power data includes: performing mutual information calculation on the historical meteorological data and the historical power data to obtain the joint probability distribution and marginal probability distribution of each meteorological factor contained in the historical meteorological data relative to the historical power data; calculating the information entropy of each meteorological factor relative to the historical power data based on the joint probability distribution and the marginal probability distribution; determining the influence weight of each historical meteorological data relative to the historical power data based on the information entropy; wherein the influence weight is proportional to the information entropy.

[0050] In this embodiment, by calculating the joint probability distribution and marginal probability distribution of each meteorological factor relative to the historical power data, the information entropy is further calculated through the joint probability distribution and marginal probability distribution, and finally the influence weight of each historical meteorological data relative to the historical power data is determined through the information entropy.

[0051] In one embodiment, after obtaining the information entropy, an influence weight is assigned to each historical meteorological data based on the information entropy.

[0052] In one embodiment, after assigning the impact weight, the historical meteorological data is further screened based on the information entropy to obtain target meteorological data whose information entropy is greater than a preset value, and a training data set is further constructed based on the target meteorological data so that the model training for power prediction is performed considering the target meteorological data that is strongly correlated with the power data.

[0053] Based on this, as an exemplary embodiment, before inputting the historical meteorological data, historical power data, the reconstructed intrinsic mode function and the residual term into a pre-built preset model, the method for constructing a power prediction model based on mutual information and signal decomposition also includes: screening the historical meteorological data based on the information entropy to obtain target meteorological data; wherein the information entropy of each meteorological factor contained in each target meteorological data is greater than a first preset value; and constructing a training data set based on the target meteorological data, the historical power data, the reconstructed intrinsic mode function and the residual term.

[0054] As an exemplary embodiment, the historical power data is subjected to complete empirical mode decomposition to obtain a signal decomposition result, including: adding multiple Gaussian white noises to the historical power data to construct multiple sub-historical power data sequences; performing EMD decomposition on each of the sub-historical power data sequences 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 of the sub-historical power data sequences, and the residual sequence is obtained by taking the difference between each of the sub-historical power data sequences and the initial intrinsic mode function; performing EMD on the residual sequence Perform multiple rounds of EMD decomposition, and in each round of EMD decomposition, add multiple Gaussian white noises to the residual sequence obtained in the previous round of decomposition to obtain multiple sub-residual sequences; perform EMD decomposition on each of the sub-residual sequences to obtain the iterative intrinsic mode function of the current round and the residual sequence of the current round, until a preset number of iterations is reached, and use the residual sequence obtained in the last round of EMD decomposition as the residual term, and use the iterative intrinsic mode function and the initial intrinsic mode function obtained in each round of EMD decomposition as the intrinsic mode function to obtain the signal decomposition result.

[0055] Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) is a signal processing technique used to decompose complex time series data into a series of simple intrinsic mode functions (IMFs) with different frequencies and a residual term. CEEMDAN is an extension of empirical mode decomposition (EMD) that improves the accuracy and reliability of the decomposition by introducing an ensemble averaging process and adaptive noise.

[0056] Specifically, in this embodiment, for the complete empirical mode decomposition of historical power data, multiple Gaussian white noises are first added to the historical power data to construct multiple sub-historical power data sequences; further, each of the sub-historical power data sequences is subjected to EMD decomposition 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 of the sub-historical power data sequences, and the residual sequence is obtained by taking the difference between each of the sub-historical power data sequences and the initial intrinsic mode function.

[0057] 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 of the sub-residual sequences to obtain the iterative intrinsic mode function of the current round and the residual sequence of the current round until a 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 to obtain the signal decomposition result.

[0058] As an exemplary embodiment, the preset model includes a first feature extraction module, a second feature extraction module and a power output module, and the historical meteorological data, the historical power data, the reconstructed intrinsic mode function and the residual term are input into the pre-built preset model, and the model training is performed under the constraint of the influence weight; including: inputting the historical meteorological data, the reconstructed intrinsic mode function and the residual term into the preset model for model training, continuously adjusting the parameters of the first feature extraction module so that the first feature extraction module can learn the correspondence between the historical meteorological data and the reconstructed intrinsic mode function and the residual term under the influence weight; continuously adjusting the parameters of the second feature extraction module so that the second feature extraction module can learn the correspondence between the local time series change features and the global time series change features extracted from the reconstructed intrinsic mode function and the residual term; continuously adjusting the parameters of the power output module so that the power output module learns the correspondence between the local time series change features and the global time series change features and the historical power data.

[0059] In this embodiment, the preset model includes a first feature extraction module, a second feature extraction module and a power output module.

[0060] Exemplarily, the first feature extraction module is used to learn the correspondence between the historical meteorological data and the reconstructed intrinsic mode function and the residual term under the influence weight; the setting and training method of the first feature extraction module, the power prediction model finally obtained can obtain the reconstructed intrinsic mode function and residual term corresponding to the meteorological data by inputting meteorological data, and can determine the reconstructed intrinsic mode function and residual term of the power data corresponding to the contribution of each meteorological factor contained in the meteorological data to the power prediction, so as to fully consider the complexity, nonlinearity and non-stationarity between historical meteorological data and power data; the reconstructed intrinsic mode function can represent power fluctuations on different time scales, which includes the power fluctuation state under extreme weather conditions, and the residual term can reflect the characteristic performance of historical power data on a long time scale. Therefore, the setting of the first feature extraction module can consider the fluctuation characteristics of the power data corresponding to the input meteorological data when predicting power.

[0061] In one embodiment, the first feature extraction module may be a convolutional neural network (CNN) or a recurrent neural network (RNN).

[0062] Exemplarily, the second feature extraction module is used to learn the correspondence between the local time series change features and the global time series change features extracted from the reconstructed intrinsic mode function and the residual term, so that the power prediction model finally trained can determine the local time series change features and the global time series change features corresponding to the reconstructed intrinsic mode function and the residual term corresponding to the input meteorological data.

[0063] Exemplarily, the second feature extraction module may be at least one of an Informer and a Transformer.

[0064] Exemplarily, the power output module is used to learn the corresponding relationship between the local time series change feature, the global time series change feature and the historical power data.

[0065] The above-mentioned method of this embodiment, the preset model includes a first feature extraction module, a second feature extraction module and a power output module. The setting and training method of the first feature extraction module, the power prediction model finally obtained can obtain the reconstructed intrinsic mode function and residual term corresponding to the meteorological data by inputting meteorological data, and can determine the reconstructed intrinsic mode function and residual term of the power data corresponding to the meteorological factors contained in the meteorological data based on the contribution of the meteorological factors to the power prediction, so as to fully consider the complexity, nonlinearity and non-stationarity between the historical meteorological data and the power data; the reconstructed intrinsic mode function can represent the power fluctuation on different time scales, which includes the power fluctuation state under extreme weather conditions, and the residual term can reflect the characteristic performance of the historical power data on a long time scale. Therefore, the setting of the first feature extraction module can consider the fluctuation characteristics of the power data corresponding to the input meteorological data when predicting power; the setting of the second feature extraction module and the power prediction module can fully consider the global time series characteristics and local time series characteristics of the reconstructed intrinsic mode function and the residual term for power prediction, so that the model can capture local information while also capturing global information well, and can consider the power fluctuation condition under extreme weather conditions for power prediction, thereby improving the prediction accuracy.

[0066] As an exemplary embodiment, the preset model is the CNTTansNet model; this model combines the CNN and Time-Series Transformer methods, and changes the attention mechanism inside the Transformer to a depth-separable convolution, so that it can capture local and global context information at the same time, thereby improving prediction accuracy.

[0067] Depthwise Separable Convolution (DSC) is an efficient CNN structure that decomposes the standard convolution operation into two smaller operations: depthwise convolution (DC) and pointwise convolution (PC). Compared with standard convolution, DSC greatly reduces the number of model parameters and computational complexity. In DC, since each channel is convolved independently, the number of parameters is proportional to the number of input channels. In PC, the number of parameters is proportional to the product of the number of input and output channels, which is usually much smaller than the number of parameters of standard convolution.

[0068] Among them, the CNTTansNet model uses one-dimensional convolution operations to extract local features of the data, and replaces the traditional multi-head attention mechanism with depthwise separable convolution inside the Transformer encoder; this design not only reduces the computational complexity, but also can more effectively capture the global dependencies in time series data.

[0069] Exemplarily, the model structure of the CNTTansNet model includes an input layer, a CNN layer, a depth-wise separable convolution layer, a Transformer encoder layer and an output layer; in the training stage, the model continuously optimizes parameters through forward propagation and back propagation, so that CNTTansNet can learn the corresponding relationship between the reconstructed intrinsic mode function of the historical meteorological data, the residual term and the historical power data, with minimizing the mean square error between the predicted value and the true value as the optimization goal; in the prediction stage, the preprocessed input data undergoes multi-layer abstraction and feature extraction of the CNTTansNet model, and finally outputs the predicted wind power value; in the above embodiment, the CNTTansNet model can determine the reconstructed intrinsic mode function and residual term of the corresponding power data based on the contribution of each meteorological factor contained in the meteorological data to the power prediction, so as to fully consider the relationship between the historical meteorological data and the power data. Complexity, nonlinearity and non-stationarity; the reconstructed intrinsic mode function can represent the power fluctuations on different time scales, which includes the power fluctuation state under extreme weather conditions, and the residual term can reflect the characteristic performance of historical power data on a long time scale. Therefore, the setting of the first feature extraction module can consider the fluctuation characteristics of the power data corresponding to the input meteorological data when predicting power; the setting of the second feature extraction module and the power prediction module can fully consider the global time series characteristics and local time series characteristics of the reconstructed intrinsic mode function and the residual term for power prediction, so that the model can capture local information while also capturing global information well, and can consider the power fluctuation conditions under extreme weather conditions for power prediction, thereby improving the prediction accuracy; experimental results show that the CNTTansNet model performs well in both computational efficiency and prediction accuracy, and can provide an efficient and accurate solution for short-term prediction of wind power.

[0070] Example 2: This embodiment provides a power prediction model construction device based on mutual information and signal decomposition, such as Figure 2 Shown, including:

[0071] The acquisition module 501 is used to acquire historical meteorological data and historical power data of the wind farm station; wherein the historical meteorological data includes time series data corresponding to multiple types of meteorological factors;

[0072] A calculation module 502 is configured to perform mutual information calculation on the historical meteorological data and the historical power data to obtain an influence weight of each historical meteorological data relative to the historical power data;

[0073] A signal decomposition module 503 is configured to perform complete empirical mode decomposition on the historical power data to obtain a signal decomposition result; wherein the signal decomposition result includes a plurality of intrinsic mode functions with different frequencies and a residual term corresponding to the signal decomposition result; each of the intrinsic mode functions is used to characterize the historical fluctuations of the historical power data at different time scales, and the residual term is used to characterize the overall trend of the historical power data;

[0074] A feature reconstruction module 504 is used to perform feature screening and feature reconstruction based on the signal decomposition result to obtain a reconstructed intrinsic mode function and a residual term;

[0075] The model training module 505 is used to input the historical meteorological data, historical power data and the reconstructed intrinsic mode function into a pre-built preset model, and perform model training under the constraints of the influence weight until the model converges to obtain a power prediction model; during the model training process, the correspondence between the historical meteorological data and the reconstructed intrinsic mode function, the residual term and the historical power data is learned.

[0076] It should be noted here that the examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the contents disclosed in the above embodiments.

[0077] It should be noted that the above modules as part of the device can be implemented through software or hardware, wherein the hardware environment includes a network environment.

[0078] Example 3. An embodiment of the present invention further provides a computer device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus, the memory is used to store computer programs; the processor is used to execute the method in any of the above embodiments by running the computer program stored in the memory.

[0079] Figure 3 is a structural block diagram of an optional computer device according to an embodiment of the present application, such as Figure 3 As shown, it includes a processor 10, a communication interface 20, a memory 30 and a communication bus 40, wherein the processor 10, the communication interface 20 and the memory 30 communicate with each other through the communication bus 40, wherein,

[0080] Memory 30, for storing computer programs;

[0081] The processor 10 is configured to implement the method of any of the above embodiments when executing the computer program stored in the memory 30 .

[0082] Optionally, in this embodiment, the communication bus may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The communication bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 3 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0083] The communication interface is used for communication between the above-mentioned computer device and other devices.

[0084] The memory may include RAM, or may include non-volatile memory, such as at least one disk memory. Alternatively, the memory may also be at least one storage device located away from the aforementioned processor.

[0085] The above-mentioned processor can be a general-purpose processor, which can include but is not limited to: CPU (Central Processing Unit), NP (Network Processor), etc.; it can also be DSP (Digital Signal Processing), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0086] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments, and this embodiment will not be described in detail here.

[0087] It can be understood by those skilled in the art that Figure 3 The structure shown is for illustration only. The device for implementing any one of the methods in the above embodiments may be a terminal device, which may be a smart phone (such as an Android phone, an IOS phone, etc.), a tablet computer, a PDA, a mobile Internet device (MID), a PAD, or other terminal devices. Figure 3 It does not limit the structure of the above electronic device. For example, the terminal device may also include Figure 3 More or fewer components (such as network interfaces, display devices, etc.) shown in, or with Figure 3 Different configurations shown.

[0088] A person skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, which can include: a flash drive, ROM, RAM, a magnetic disk or an optical disk, etc.

[0089] As an exemplary embodiment, the present application further provides a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute any one of the method steps of the present embodiment when run.

[0090] Optionally, in this embodiment, the above-mentioned storage medium can be used to execute the program code of the method steps of the embodiment of the present application.

[0091] Optionally, in this embodiment, the above-mentioned storage medium may be located on at least one network device among the multiple network devices in the network shown in the above-mentioned embodiment.

[0092] Optionally, in this embodiment, the storage medium is configured to store data for executing the method in the above embodiment.

[0093] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments, which will not be described in detail in this embodiment.

[0094] Optionally, in this embodiment, the storage medium may include, but is not limited to, various media that can store program codes, such as a USB flash drive, a ROM, a RAM, a mobile hard disk, a magnetic disk, or an optical disk.

[0095] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0096] If the integrated units in the above embodiments are implemented in the form of software functional units and sold or used as independent products, they can be stored in the above-mentioned computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing one or more computer devices (which may be personal computers, servers, or network devices, etc.) to execute all or part of the steps of the method in the above embodiments.

[0097] In the several embodiments provided in this application, it should be understood that the disclosed client can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is merely a logical functional division. In actual implementation, there may be other division methods, such as combining or integrating multiple units or components into another system, or ignoring or not implementing some features. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interface, indirect coupling or communication connection of units or modules, and may be electrical or other forms.

[0098] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected based on actual needs to achieve the purpose of the solution provided in this embodiment.

[0099] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0100] In the above embodiments of the present application, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0101] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for constructing a power prediction model based on mutual information and signal decomposition, characterized by: include, Acquire historical meteorological data and historical power data of the wind farm station; wherein the historical meteorological data includes time series data corresponding to multiple types of meteorological factors; Performing mutual information calculation on the historical meteorological data and the historical power data to obtain an influence weight of each of the historical meteorological data relative to the historical power data; Performing complete empirical mode decomposition on the historical power data to obtain a signal decomposition result; wherein the signal decomposition result includes a plurality of intrinsic mode functions with different frequencies and a residual term corresponding to the signal decomposition result; each of the intrinsic mode functions is used to characterize the historical fluctuations of the historical power data at different time scales, and the residual term is used to characterize the overall trend of the historical power data; Performing feature screening and feature reconstruction based on the signal decomposition result to obtain a reconstructed intrinsic mode function and a residual term; The historical meteorological data, the historical power data, the reconstructed intrinsic mode function and the residual term are input into a pre-built preset model, and the model is trained under the constraint of the influence weight until the model converges to obtain a power prediction model; during the model training process, the correspondence between the historical meteorological data and the reconstructed intrinsic mode function, the residual term and the historical power data is learned.

2. The method for constructing a power prediction model based on mutual information and signal decomposition according to claim 1, wherein: The performing mutual information calculation on the historical meteorological data and the historical power data to obtain the influence weight of each historical meteorological data relative to the historical power data includes: Performing mutual information calculation on the historical meteorological data and the historical power data to obtain a joint probability distribution and a marginal probability distribution of each meteorological factor included in the historical meteorological data relative to the historical power data; Calculating the information entropy of each meteorological factor relative to the historical power data based on the joint probability distribution and the marginal probability distribution; Based on the information entropy, the influence weight of each of the historical meteorological data relative to the historical power data is determined; wherein the influence weight is proportional to the information entropy.

3. The method for constructing a power prediction model based on mutual information and signal decomposition according to claim 2, wherein: Before inputting the historical meteorological data, the historical power data, the reconstructed intrinsic mode function and the residual term into the pre-built preset model, the power prediction model construction method based on mutual information and signal decomposition further includes: The historical meteorological data are screened based on the information entropy to obtain target meteorological data; wherein the information entropy of each meteorological factor included in each target meteorological data is greater than a first preset value; A training data set is constructed based on the target meteorological data, the historical power data, the reconstructed intrinsic mode function and the residual term.

4. The method for constructing a power prediction model based on mutual information and signal decomposition according to claim 3, wherein: The performing complete empirical mode decomposition on the historical power data to obtain a signal decomposition result includes: Adding multiple Gaussian white noises to the historical power data to construct multiple sub-historical power data sequences; Performing EMD decomposition on each of the sub-historical power data sequences 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 of the sub-historical power data sequences, and the residual sequence is obtained by subtracting each of the sub-historical power data sequences from the initial intrinsic mode function; 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 in the previous round of decomposition to obtain multiple sub-residual sequences; EMD decomposition is performed on each of the sub-residual sequences to obtain the iterative intrinsic mode function of the current round and the residual sequence of the current round, until a preset number of iterations is reached, the residual sequence obtained in the last round of EMD decomposition is used as the residual term, 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, and the signal decomposition result is obtained.

5. The method for constructing a power prediction model based on mutual information and signal decomposition according to claim 4, wherein: The preset power prediction model includes a first feature extraction module, a second feature extraction module, and a power output module. The historical meteorological data, the historical power data, the reconstructed intrinsic mode function, and the residual term are input into a pre-built preset model, and the model is trained under the constraint of the influence weight; including: Inputting the historical meteorological data, the reconstructed intrinsic mode function, and the residual term into the preset power prediction model for model training, and continuously adjusting the parameters of the first feature extraction module so that the first feature extraction module can learn the corresponding relationship between the historical meteorological data and the reconstructed intrinsic mode function and the residual term under the influence weight; Continuously adjusting the parameters of the second feature extraction module so that the second feature extraction module can learn the correspondence between the local time series change feature and the global time series change feature extracted from the reconstructed intrinsic mode function and the residual term; Parameters of the power output module are continuously adjusted so that the power output module learns the corresponding relationship between the local time series variation characteristics, the global time series variation characteristics, and the historical power data.

6. The method for constructing a power prediction model based on mutual information and signal decomposition according to claim 5, wherein: The preset power prediction model is the CNTTansNet model.

7. The method for constructing a power prediction model based on mutual information and signal decomposition according to claim 6, wherein: The first feature extraction module includes at least one of CNN and RNN, and the second feature extraction module includes at least one of Informer and Transformer.

8. A power prediction model construction device based on mutual information and signal decomposition, applying the power prediction model construction method based on mutual information and signal decomposition according to any one of claims 1 to 7, characterized in that: include: An acquisition module is used to acquire historical meteorological data and historical power data of a wind farm station; wherein the historical meteorological data includes time series data corresponding to multiple types of meteorological factors; a calculation module, configured to perform mutual information calculation on the historical meteorological data and the historical power data to obtain an influence weight of each of the historical meteorological data relative to the historical power data; A signal decomposition module, configured to perform complete empirical mode decomposition on the historical power data to obtain a signal decomposition result; wherein the signal decomposition result includes a plurality of intrinsic mode functions with different frequencies and a residual term corresponding to the signal decomposition result; each of the intrinsic mode functions is used to characterize the historical fluctuations of the historical power data at different time scales, and the residual term is used to characterize the overall trend of the historical power data; A feature reconstruction module is used to perform feature screening and feature reconstruction based on the signal decomposition result to obtain a reconstructed intrinsic mode function and a residual term; A model training module is used to input the historical meteorological data, historical power data and the reconstructed intrinsic mode function into a pre-built preset model, and perform model training under the constraints of the influence weights until the model converges to obtain a power prediction model; during the model training process, the correspondence between the historical meteorological data and the reconstructed intrinsic mode function, the residual term and the historical power data is learned.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for constructing a power prediction model based on mutual information and signal decomposition according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for constructing a power prediction model based on mutual information and signal decomposition according to any one of claims 1 to 7 are implemented.

Images