An optimization method for controlling wind power prediction error during black start of overcapacity energy storage

The wind power prediction error is hierarchically processed through the layered Bayesian model and the Markov chain Monte Carlo method, and combined with multimodal data and multi-scale analysis, the shortcomings of the wind power prediction model in error distribution and wind condition changes are solved. Improve prediction accuracy and stability and reduce the impact on grid operation.

CN118693887BActive Publication Date: 2024-11-29XIAN THERMAL POWER RES INST CO LTD +1
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202411173254.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-26
Publication Date
2024-11-29
Estimated Expiration
2044-08-26

AI Technical Summary

Technical Problem

The existing wind power prediction model is difficult to deal with multi-level error distribution, resulting in inaccurate prediction results in different time periods and wind conditions, affecting the stable operation of the power grid and power scheduling.

Method used

The hierarchical Bayesian model is used to combine the Markov chain Monte Carlo method or variational inference method to hierarchically process and correct the prediction errors. Through multimodal data preprocessing and multi-scale time series analysis, short-term and medium-term and long-term prediction models, and use twin networks for similarity learning to optimize prediction results.

Benefits of technology

It improves the accuracy and stability of wind power prediction, reduces the impact of prediction errors on grid operation and power scheduling, and enhances the adaptability and flexibility of the model.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118693887B_ABST
    Figure CN118693887B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for optimizing the control of wind power prediction error during black start of super-capacity energy storage, which relates to the technical field of black start of super-capacity energy storage, including collecting multimodal data, preprocessing the multimodal data, and inputting the preprocessed multimodal data into a multimodal variational autoencoder model for data fusion to generate a unified feature representation; using the fused data to build a prediction model, and combining wavelet transform to perform multi-scale time series analysis; building a twin network for similarity learning, inputting historical data sample pairs, outputting similarity scores, identifying similar wind conditions in historical data, and optimizing the prediction model; establishing a hierarchical Bayesian model, processing and correcting the prediction error, and solving it to obtain the corrected prediction result. The present invention realizes hierarchical processing and correction of wind power prediction errors through a hierarchical Bayesian model and an advanced solution method.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of overcapacity energy storage black start, and in particular to an overcapacity energy storage black start wind power prediction error control optimization method. Background Art

[0002] Wind power forecasting system is an important part of modern power system. With the continuous increase of wind power installed capacity, accurate wind power forecasting is of great significance for the stable operation of power grid and power dispatching. However, there are many challenges in wind power forecasting, including the high uncertainty and complexity of meteorological factors such as wind speed and wind direction. These factors lead to large errors in forecasting models in practical applications. Traditional forecasting models are difficult to take into account changes in different time scales at the same time, and when facing different wind conditions, the forecast error distribution is uneven, which affects the accuracy and reliability of the forecast results. Therefore, it is an urgent problem to establish a forecasting model that can handle multi-level error distribution and can be corrected in different time periods and wind conditions. Summary of the invention

[0003] In view of the above existing problems, the present invention is proposed.

[0004] Therefore, the present invention aims to solve the deficiencies of existing wind power prediction models in dealing with multi-level error distributions, and to provide a method that can hierarchically process and correct prediction errors. Specifically, the present invention establishes a hierarchical Bayesian model to hierarchically process the overall error distribution, the error distribution in different time periods, and the error distribution under specific wind conditions for the prediction error, and uses the Markov chain Monte Carlo method or the variational inference method to solve the model, thereby obtaining the corrected prediction results. Through this method, the accuracy and stability of wind power prediction can be effectively improved, and the impact of prediction errors on power grid operation and power dispatch can be reduced.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions, a method for optimizing the control of wind power prediction error of black start of ultra-capacity energy storage, comprising:

[0006] Collect multimodal data, perform multimodal data preprocessing, and input the preprocessed multimodal data into the multimodal variational autoencoder model for data fusion to generate a unified feature representation; use the fused data to build prediction models on short-term, medium-term and long-term time scales, and combine wavelet transform to perform multi-scale time series analysis; build a twin network for similarity learning, input historical data sample pairs, output similarity scores, identify similar wind conditions in historical data, and optimize the prediction model; establish a hierarchical Bayesian model, perform hierarchical processing and correction on the prediction error, use the Markov chain Monte Carlo method or variational inference method to solve the hierarchical Bayesian model, and obtain the corrected prediction results.

[0007] As a preferred solution of the ultra-capacity energy storage black start wind power prediction error control optimization method described in the present invention, wherein: the collected multimodal data includes meteorological data, historical wind power data and environmental sensor data; the meteorological data includes wind speed, wind direction, temperature, and humidity; the historical wind power data includes power generation power and wind speed records; the environmental sensor data includes temperature sensor and humidity sensor data; the multimodal data preprocessing includes data normalization, denoising and filling missing values, unifying the data format and cleaning the data; wherein the data normalization process converts data of different dimensions into the same dimension, the denoising process uses a low-pass filter and a mean filter, and the missing values ​​are filled using interpolation and filling methods.

[0008] As a preferred solution of the ultra-capacity energy storage black start wind power prediction error control optimization method described in the present invention, wherein: the multimodal variational autoencoder model includes an encoder, a decoder and a latent space; the encoder maps the multimodal data to the latent space, and the decoder restores the representation of the latent space to the original data;

[0009] The generating of a unified feature representation includes constructing a multimodal variational autoencoder model data fusion formula as follows:

[0010] ; Information filtering function It can be expressed as the operation process of a multi-layer convolutional neural network, specifically:

[0011] ;

[0012] in, is the fused feature representation, is the number of data modes, is the data index of the modality, For input data, is the data weight matrix of the i-th mode, is the activation function, is a complex information filtering function, is the mean of the jth data in the ith mode, is the standard deviation of the jth data in the ith mode, is the data dimension for each modality; is the time variable, used for integration operation; is the partial derivative with respect to time, is the integration variable of the data, is the number of convolution kernels, is the feature after convolution, is the mean vector, is the covariance matrix, is the number of convolutional layers; For the The convolution kernel of the layer; For the The input features of the layer; For the Bias of the layer; Represents a convolution operation.

[0013] As a preferred solution of the ultra-capacity energy storage black start wind power prediction error control optimization method described in the present invention, wherein: the constructed prediction model includes a short-term prediction model, a medium-term prediction model and a long-term prediction model, the short-term prediction model adopts the long short-term memory network LSTM, the medium-term prediction model adopts the autoregressive integral moving average model ARIMA, and the long-term prediction model adopts the trend analysis method, and combines the wavelet transform to perform multi-scale time series analysis as follows:

[0014] ;in, is the prediction result after multi-scale time series analysis, is the time scale quantity, is the sth time scale, is the time range of the sth time scale, is the number of data points in the sth time scale, is the weight matrix of the sth time scale, is the activation function, is the information filtering function of the s-th time scale, is the convolution kernel of the sth time scale, For input data, is the mean of the ith data point in the sth time scale, is the covariance matrix of the ith data point in the sth time scale, is the trend analysis function of the sth time scale, is the data point at the sth time scale, is the integral variable of the sth time scale; the outer sum Indicates the sum of the prediction results of all time scales to ensure that the information of all time scales is taken into account; inner summation Indicates that all data points are summed at each time scale and the information of all data points is integrated; the weight matrix Represents the weight at each time scale; exponential function Represents the Gaussian distribution of data, used to add nonlinear transformations in the feature extraction process; activation function ; Represents the operation process of a multi-layer convolutional neural network to extract information from the data, where is the number of convolutional layers, is the convolution kernel of the lth layer at the sth time scale, For the The input features of the layer, is the bias of the lth layer at the sth time scale; Indicates trend analysis at each time scale, where and is the parameter of the linear trend; the specific expansion is:

[0015] .

[0016] As a preferred solution of the super-capacity energy storage black start wind power prediction error control optimization method described in the present invention, wherein: the construction of the prediction model includes analyzing and judging according to the prediction results, assuming that the time span is T, the short-term, medium-term and long-term time windows are ;

[0017] Short-term volatility calculation:

[0018] ,

[0019] in, It is the average of the short-term forecast results. Medium-term trend slope calculation:

[0020] ,

[0021] in, is the average value of time in the medium-term time window, It is the average of the medium-term forecast results. Long-term trend slope calculation:

[0022] ,

[0023] in, is the average value of time in the long-term time window, is the average of the forecast results in the long term.

[0024] If the short-term volatility σ short Greater than the threshold σ threshold , and the frequency of fluctuations exceeds expectations, then the forecast results reflect short-term fluctuations;

[0025] like Greater than threshold , then the forecast results reflect an obvious medium-term trend;

[0026] like Greater than threshold , and Less than threshold , then the forecast results reflect the long-term trend;

[0027] If σ short , β medium and Equilibrium, that is, σ short , β medium and If it is within the expected range, the prediction result reflects the balance of multi-scale information.

[0028] As a preferred solution of the method for optimizing the prediction error control of wind power from black start of super capacity energy storage described in the present invention, the analysis and judgment based on the prediction results include: if the short-term volatility σ short Greater than the threshold σ threshold , and the frequency of fluctuations exceeds expectations, real-time data monitoring is continued, and the number of sensors and the frequency of data collection are increased; when frequency fluctuations beyond expectations are detected, the data collection frequency is increased to every minute or every second, and a rapid response mechanism is established, including automatic adjustment of system parameters and manual intervention; when the predicted fluctuation exceeds the preset threshold, the output power of the wind turbine and the charge and discharge rate of the super-capacity energy storage system are automatically adjusted, and the operation and maintenance personnel are notified to conduct on-site inspections; if short-term fluctuations continue to intensify, the short-term prediction model LSTM is optimized, and the responsiveness of the model is improved by introducing new real-time data and repeated training; when the optimization is completed, the new model is deployed to the real-time monitoring system, and the model performance is regularly evaluated and adjusted; if the short-term fluctuation still exceeds the threshold, the alarm system is triggered, and a multi-level alarm mechanism is set to notify operation and maintenance personnel at different levels according to the degree of fluctuation; when the alarm system is started, the corresponding data is automatically recorded, and a fluctuation analysis report is provided for reference by the operation and maintenance personnel.

[0029] If the medium-term trend slope β medium Greater than the threshold β medium-threshold , and the standard deviation in the medium term is medium Less than the threshold σ medium-threshold , a medium-term trend tracking system is established, including a trend analysis module and a report generation module; when a medium-term trend is detected, the trend changes are continuously monitored, a medium-term trend analysis report is generated, updated weekly or monthly, and the operating parameters of the medium-term trend tracking system are adjusted according to the trend changes, and the charging and discharging strategy of the energy storage system and the operating mode of the wind turbine are adjusted; when the adjustment is completed, the new parameters are applied for operation, and a regular parameter inspection and adjustment mechanism is set up. At the same time, the medium-term prediction model ARIMA is optimized to improve the prediction accuracy by adding historical data and improving the model structure; when the model optimization is completed, the model is verified and the performance is evaluated, and it is deployed to the corresponding equipment for real-time prediction, the medium-term operation strategy is optimized, and the maintenance plan and resource allocation are adjusted; when the strategy adjustment is completed, the new strategy is implemented.

[0030] If the long-term trend slope β long Greater than the threshold β long-threshold, and the standard deviation in the long term σ long Less than the threshold σ long-threshold , then formulate long-term plans and strategies, including equipment updates, technology improvements and resource allocation, and adjust policies and operating strategies according to trends, and formulate new capacity planning and energy storage system configuration plans; when policy adjustments are completed, implement new capacity planning and energy storage system configuration plans, regularly update long-term prediction models by introducing historical data and improving model structures, conduct model verification and performance evaluation, and then deploy them to the corresponding energy storage systems for long-term prediction.

[0031] If the short-term volatility σ short , medium-term trend slope β medium and the long-term trend slope β long If there is no single scale dominating the equilibrium, a comprehensive monitoring system will be established, including short-term, medium-term and long-term monitoring modules; when equilibrium information is detected, short-term, medium-term and long-term changes will be monitored simultaneously, a comprehensive analysis report will be generated, and the short-term, medium-term and long-term prediction models will be optimized by introducing multi-scale data and improving the model structure; when the model optimization is completed, the model verification and performance evaluation will be carried out, and then it will be deployed to the comprehensive monitoring system for comprehensive prediction, and a coordinated strategy will be formulated to ensure that the optimization measures at each time scale cooperate with each other; when the strategy is formulated, the coordinated strategy will be implemented, and at the same time, the system operating parameters will be flexibly adjusted according to the multi-scale information, and the charging and discharging strategy of the energy storage system and the operating mode of the wind turbine will be adjusted; when the adjustment is completed, the optimal operation plan will be implemented.

[0032] That is, if the forecast results reflect short-term fluctuations, real-time data monitoring and rapid response mechanisms will be continuously carried out, system parameters will be adjusted and short-term forecast models will be optimized; when short-term fluctuations are significant, a multi-level alarm system will be triggered and data will be recorded; if the forecast results reflect obvious medium-term trends, a trend tracking system should be established, system parameters should be adjusted and medium-term forecast models should be optimized, and the optimized medium-term operation strategy should be implemented; if the forecast results reflect long-term change trends, long-term plans and strategies should be formulated, policies and operation strategies should be adjusted, long-term forecast models should be updated regularly, and investment plans should be formulated; if the forecast results reflect multi-scale information balance, a comprehensive monitoring system will be established, and the forecast models of each time scale will be optimized at the same time, coordination strategies will be formulated and system parameters will be flexibly adjusted, and the optimal operation plan will be implemented to ensure the stability and efficiency of the system operation.

[0033] As a preferred solution of the ultra-capacity energy storage black start wind power prediction error control optimization method described in the present invention, wherein: the construction of the twin network for similarity learning includes, the twin network is composed of two identical neural networks, two different samples are input, and similarity is output. The twin network is trained by contrastive learning, and positive sample pairs and negative sample pairs are input to minimize the distance of the positive sample pairs and maximize the distance of the negative sample pairs:

[0034] in, is the loss function, represents a positive sample pair, represents a negative sample pair, is the time range, is the number of data points, is the partial derivative with respect to time, For sample The characteristic vector of For the The similarity score of each data point.

[0035] The core of the formula is to optimize the twin network by calculating the loss function, which is composed of the sum of the losses of positive sample pairs and negative sample pairs.

[0036] The loss of positive sample pairs is achieved by minimizing the distance between two similar samples. Sum, calculate the time integral and nested sum of the feature vector difference; the loss of negative sample pairs is achieved by maximizing the distance between two dissimilar samples. Perform summation, compute time integrals of eigenvector differences, and nested summations.

[0037] The core of the twin network is to calculate the feature vector of the input sample ,The difference of the feature vector is calculated by the partial derivative of time, reflecting the changing trend among the samples.

[0038] Activation Function ; The information filtering function is ,in is the filtering parameter, a is the input;

[0039] The final expansion formula is:

[0040] ,

[0041] in, is the value of the loss function, whose range is a set of non-negative real numbers ; The smaller the value, the higher the similarity between the positive sample pairs, the greater the difference between the negative sample pairs, and the better the network's recognition effect; the larger the value, the worse the network's recognition effect.

[0042] As a preferred solution of the ultra-capacity energy storage black start wind power prediction error control optimization method described in the present invention, wherein: the hierarchical Bayesian model includes a multi-level error distribution, the hierarchical structure includes the overall error distribution, the error distribution of different time periods and the error distribution under the marked wind conditions, and the model is solved by using the Markov chain Monte Carlo method or the variational reasoning method. The Markov chain Monte Carlo method includes using random sampling technology to estimate the parameters of the Bayesian model, and optimizing the variational lower bound by the variational reasoning method, and approximate reasoning of the Bayesian model. The formula is as follows:

[0043] ,

[0044] in, is the corrected prediction result, is the number of data samples, For the The time range of samples, y is the error vector, For the The mean of the samples, For the The covariance matrix of samples is is the dimension of the error vector, is a hidden variable, is the parameter vector of the Bayesian model, is the latent variable distribution under given parameters, is the prior distribution of the parameter. The error distribution is assumed to be Gaussian with the mean parameter and the covariance matrix ; Gaussian distribution represents the error vector Gaussian distribution; Gaussian distribution integral represents the distribution characteristics of the error vector, and the integral of latent variables and parameters represents the hierarchical structure and reasoning process of the Bayesian model. and parameters The prior distribution and conditional distribution of are modeled; Markov Chain Monte Carlo (MCMC) is used to solve the hierarchical Bayesian model and estimate the parameters of the Bayesian model through random sampling techniques.

[0045] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, steps of a method for optimizing control of wind power prediction errors during black start of an overcapacity energy storage are implemented.

[0046] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a method for optimizing wind power prediction error control during black start of an ultra-capacity energy storage system.

[0047] Beneficial effects of the present invention: The wind power prediction error control optimization method provided by the present invention has significant beneficial effects. First, by establishing a hierarchical Bayesian model, the prediction error is hierarchically processed, including the overall error distribution, the error distribution in different time periods, and the error distribution under specific wind conditions, so that the prediction model can comprehensively consider the different sources and characteristics of the error. Secondly, the Markov chain Monte Carlo method or the variational inference method is used to solve the hierarchical Bayesian model, which can accurately estimate the parameters of the Bayesian model and achieve accurate error correction, thereby improving the accuracy of the prediction results. In addition, the hierarchical Bayesian model can adapt to changes in different time scales and wind conditions, has strong flexibility and adaptability, and can be adjusted and optimized according to different wind farms and meteorological conditions in practical applications. Through effective error correction and optimization of prediction results, the impact of wind power prediction errors on power grid operation and power dispatching is reduced, and the stability and reliability of the power system are improved. The method of the present invention can process a large number of historical data samples, and through the combination of the hierarchical Bayesian model and the Gaussian distribution, the information of historical data is fully utilized to improve the performance of the prediction model. In summary, the present invention realizes hierarchical processing and correction of wind power prediction errors through a hierarchical Bayesian model and an advanced solution method, and provides an effective wind power prediction error control optimization method, which has significant application value and promotion prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0049] Figure 1 A schematic flow chart of a method for optimizing wind power prediction error control during black start of an ultra-capacity energy storage system is provided as an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0051] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0052] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0053] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0054] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0055] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0056] Example 1, reference Figure 1 , which is the first embodiment of the present invention, and provides an optimization method for controlling wind power prediction error during black start of an ultra-capacity energy storage system, comprising:

[0057] Collect multimodal data, perform multimodal data preprocessing, and input the preprocessed multimodal data into the multimodal variational autoencoder model for data fusion to generate a unified feature representation; use the fused data to build prediction models on short-term, medium-term and long-term time scales, and combine wavelet transform to perform multi-scale time series analysis; build a twin network for similarity learning, input historical data sample pairs, output similarity scores, identify similar wind conditions in historical data, and optimize the prediction model; establish a hierarchical Bayesian model, perform hierarchical processing and correction on the prediction error, use the Markov chain Monte Carlo method or variational inference method to solve the hierarchical Bayesian model, and obtain the corrected prediction results.

[0058] The collected multimodal data includes meteorological data, historical wind power data and environmental sensor data; the meteorological data includes wind speed, wind direction, temperature and humidity; the historical wind power data includes power generation and wind speed records; the environmental sensor data includes temperature sensor and humidity sensor data;

[0059] The multimodal data preprocessing includes data normalization, denoising and missing value completion, unifying the data format and cleaning the data; wherein the data normalization process converts data of different dimensions into the same dimension, the denoising process uses a low-pass filter and a mean filter, and the missing value completion uses an interpolation method and a filling method.

[0060] The multimodal variational autoencoder model includes an encoder, a decoder and a latent space; the encoder maps the multimodal data to the latent space, and the decoder restores the representation of the latent space to the original data;

[0061] The generating of a unified feature representation includes constructing a multimodal variational autoencoder model data fusion formula as follows:

[0062] ;

[0063] Exponential function and information filtering functions Expand it, assuming that the internal expression of the exponential function is a combination of the Gaussian function and the convolution operation, and expand it to:

[0064] ,

[0065] in, is the number of convolution kernels; is the feature after convolution; is the mean vector; is the covariance matrix. Information filtering function It can be expressed as the operation process of a multi-layer convolutional neural network, specifically:

[0066] ,

[0067] in, is the fused feature representation, is the number of data modes, is the data index of the modality, For input data, is the data weight matrix of the i-th mode, is the activation function, is a complex information filtering function, is the mean of the jth data in the ith mode, is the standard deviation of the jth data in the ith mode, is the data dimension for each modality; is the time variable, used for integration operation; is the partial derivative with respect to time, is the integration variable of the data, is the number of convolution kernels, is the feature after convolution, is the mean vector, is the covariance matrix, is the number of convolutional layers; For the The convolution kernel of the layer; For the The input features of the layer; For the Bias of the layer; Represents a convolution operation.

[0068] The constructed prediction model includes a short-term prediction model, a medium-term prediction model and a long-term prediction model. The short-term prediction model adopts the long short-term memory network LSTM, the medium-term prediction model adopts the autoregressive integrated moving average model ARIMA, and the long-term prediction model adopts the trend analysis method and combines the wavelet transform to perform multi-scale time series analysis as follows:

[0069] ;in, is the prediction result after multi-scale time series analysis, is the time scale quantity, is the sth time scale, is the time range of the sth time scale, is the number of data points in the sth time scale, is the weight matrix of the sth time scale, is the activation function, is the information filtering function of the s-th time scale, is the convolution kernel of the sth time scale, For input data, is the mean of the ith data point in the sth time scale, is the covariance matrix of the ith data point in the sth time scale, is the trend analysis function of the sth time scale, is the data point at the sth time scale, is the integral variable of the sth time scale; the outer sum Indicates the sum of the prediction results of all time scales to ensure that the information of all time scales is taken into account; inner summation Indicates that all data points are summed at each time scale and the information of all data points is integrated; the weight matrix Represents the weight at each time scale; exponential function Represents the Gaussian distribution of data, used to add nonlinear transformations in the feature extraction process; activation function ; Represents the operation process of a multi-layer convolutional neural network to extract information from the data, where is the number of convolutional layers, is the convolution kernel of the lth layer at the sth time scale, For the The input features of the layer, is the bias of the lth layer at the sth time scale; Indicates trend analysis at each time scale, where and is the parameter of the linear trend; the specific expansion is:

[0070] .

[0071] The construction of the prediction model includes analyzing and judging according to the prediction results, assuming that the time span is T, and the short-term, medium-term and long-term time windows are ;

[0072] Short-term volatility calculation:

[0073] ,

[0074] in, It is the average of the short-term forecast results. Medium-term trend slope calculation:

[0075] ,

[0076] in, is the average value of time in the medium-term time window, It is the average of the medium-term forecast results. Long-term trend slope calculation:

[0077] ,

[0078] in, is the average value of time in the long-term time window, is the average of the forecast results in the long term.

[0079] If the short-term volatility σ short Greater than the threshold σ threshold , and the frequency of fluctuations exceeds expectations, the forecast results reflect short-term fluctuations; if Greater than threshold , then the forecast results reflect an obvious medium-term trend; if Greater than threshold , and Less than threshold , the forecast results reflect the long-term trend; if σ short , β medium and Balance, that is, average short , β medium and If it is within the expected range, the prediction result reflects the balance of multi-scale information.

[0080] The analysis and judgment based on the forecast results include: if the short-term volatility σ short Greater than the threshold σ threshold , and the frequency of fluctuations exceeds expectations, real-time data monitoring is continued, and the number of sensors and the frequency of data collection are increased; when frequency fluctuations beyond expectations are detected, the data collection frequency is increased to every minute or every second, and a rapid response mechanism is established, including automatic adjustment of system parameters and manual intervention; when the predicted fluctuation exceeds the preset threshold, the output power of the wind turbine and the charge and discharge rate of the super-capacity energy storage system are automatically adjusted, and the operation and maintenance personnel are notified to conduct on-site inspections; if short-term fluctuations continue to intensify, the short-term prediction model LSTM is optimized, and the responsiveness of the model is improved by introducing new real-time data and repeated training; when the optimization is completed, the new model is deployed to the real-time monitoring system, and the model performance is regularly evaluated and adjusted; if the short-term fluctuation still exceeds the threshold, the alarm system is triggered, and a multi-level alarm mechanism is set to notify operation and maintenance personnel at different levels according to the degree of fluctuation; when the alarm system is started, the corresponding data is automatically recorded, and a fluctuation analysis report is provided for reference by the operation and maintenance personnel.

[0081] If the medium-term trend slope β medium Greater than the threshold β medium-threshold , and the standard deviation in the medium term is medium Less than the threshold σ medium-threshold, a medium-term trend tracking system is established, including a trend analysis module and a report generation module; when a medium-term trend is detected, the trend changes are continuously monitored, a medium-term trend analysis report is generated, updated weekly or monthly, and the operating parameters of the medium-term trend tracking system are adjusted according to the trend changes, and the charging and discharging strategy of the energy storage system and the operating mode of the wind turbine are adjusted; when the adjustment is completed, the new parameters are applied for operation, and a regular parameter inspection and adjustment mechanism is set up. At the same time, the medium-term prediction model ARIMA is optimized to improve the prediction accuracy by adding historical data and improving the model structure; when the model optimization is completed, the model is verified and the performance is evaluated, and it is deployed to the corresponding equipment for real-time prediction, the medium-term operation strategy is optimized, and the maintenance plan and resource allocation are adjusted; when the strategy adjustment is completed, the new strategy is implemented.

[0082] If the long-term trend slope β long Greater than the threshold β long-threshold , and the standard deviation in the long term σ long Less than the threshold σ long-threshold , then formulate long-term plans and strategies, including equipment updates, technology improvements and resource allocation, and adjust policies and operating strategies according to trends, and formulate new capacity planning and energy storage system configuration plans; when policy adjustments are completed, implement new capacity planning and energy storage system configuration plans, regularly update long-term prediction models by introducing historical data and improving model structures, conduct model verification and performance evaluation, and then deploy them to the corresponding energy storage systems for long-term prediction.

[0083] If the short-term volatility σ short , medium-term trend slope β medium and the long-term trend slope β long If there is no single scale dominating the equilibrium, a comprehensive monitoring system will be established, including short-term, medium-term and long-term monitoring modules; when equilibrium information is detected, short-term, medium-term and long-term changes will be monitored simultaneously, a comprehensive analysis report will be generated, and the short-term, medium-term and long-term prediction models will be optimized by introducing multi-scale data and improving the model structure; when the model optimization is completed, the model verification and performance evaluation will be carried out, and then it will be deployed to the comprehensive monitoring system for comprehensive prediction, and a coordinated strategy will be formulated to ensure that the optimization measures at each time scale cooperate with each other; when the strategy is formulated, the coordinated strategy will be implemented, and at the same time, the system operating parameters will be flexibly adjusted according to the multi-scale information, and the charging and discharging strategy of the energy storage system and the operating mode of the wind turbine will be adjusted; when the adjustment is completed, the optimal operation plan will be implemented.

[0084] That is, if the forecast results reflect short-term fluctuations, real-time data monitoring and rapid response mechanisms will be continuously carried out, system parameters will be adjusted and short-term forecast models will be optimized; when short-term fluctuations are significant, a multi-level alarm system will be triggered and data will be recorded; if the forecast results reflect obvious medium-term trends, a trend tracking system should be established, system parameters should be adjusted and medium-term forecast models should be optimized, and the optimized medium-term operation strategy should be implemented; if the forecast results reflect long-term change trends, long-term plans and strategies should be formulated, policies and operation strategies should be adjusted, long-term forecast models should be updated regularly, and investment plans should be formulated; if the forecast results reflect multi-scale information balance, a comprehensive monitoring system will be established, and the forecast models of each time scale will be optimized at the same time, coordination strategies will be formulated and system parameters will be flexibly adjusted, and the optimal operation plan will be implemented to ensure the stability and efficiency of the system operation.

[0085] The construction of the twin network for similarity learning includes that the twin network is composed of two identical neural networks, two different samples are input, and similarity is output. The twin network is trained by contrastive learning, and positive sample pairs and negative sample pairs are input to minimize the distance of the positive sample pairs and maximize the distance of the negative sample pairs:

[0086]

[0087] in, is the loss function, represents a positive sample pair, represents a negative sample pair, is the time range, is the number of data points, is the partial derivative with respect to time, For sample The characteristic vector of For the The similarity score of each data point.

[0088] The core of the formula is to optimize the twin network by calculating the loss function, which is composed of the sum of the losses of positive sample pairs and negative sample pairs.

[0089] The loss of positive sample pairs is achieved by minimizing the distance between two similar samples. Sum, calculate the time integral and nested sum of the feature vector difference; the loss of negative sample pairs is achieved by maximizing the distance between two dissimilar samples. Perform summation, compute time integrals of eigenvector differences, and nested summations.

[0090] The core of the twin network is to calculate the feature vector of the input sample ,The difference of the feature vector is calculated by the partial derivative of time, reflecting the changing trend among the samples.

[0091] Activation Function ; The information filtering function is ,in is the filtering parameter and a is the input.

[0092] The final expansion formula is:

[0093] ,

[0094] in, is the value of the loss function, whose range is a set of non-negative real numbers ; The smaller the value, the higher the similarity between the positive sample pairs, the greater the difference between the negative sample pairs, and the better the network's recognition effect; the larger the value, the worse the network's recognition effect.

[0095] The hierarchical Bayesian model includes a multi-level error distribution, and the hierarchical structure includes the overall error distribution, the error distribution in different time periods, and the error distribution under the marked wind conditions. The model is solved by using the Markov chain Monte Carlo method or the variational inference method. The Markov chain Monte Carlo method includes using random sampling technology to estimate the parameters of the Bayesian model, and optimizing the variational lower bound by the variational inference method to approximate the Bayesian model. The formula is as follows:

[0096] ,

[0097] in, is the corrected prediction result, is the number of data samples, For the The time range of samples, y is the error vector, For the The mean of the samples, For the The covariance matrix of samples is is the dimension of the error vector, is a hidden variable, is the parameter vector of the Bayesian model, is the latent variable distribution under given parameters, is the prior distribution of the parameter. The error distribution is assumed to be Gaussian with the mean parameter and the covariance matrix ; Gaussian distribution represents the error vector Gaussian distribution; Gaussian distribution integral represents the distribution characteristics of the error vector, and the integral of latent variables and parameters represents the hierarchical structure and reasoning process of the Bayesian model. and parameters The prior distribution and conditional distribution of are modeled; Markov Chain Monte Carlo (MCMC) is used to solve the hierarchical Bayesian model and estimate the parameters of the Bayesian model through random sampling techniques.

[0098] Embodiment 2, the second embodiment of the present invention, is different from the previous embodiment in that:

[0099] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the 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, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0100] 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 generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. 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.

[0101] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate 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 A function specified in one or more boxes.

[0102] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0103] Although the preferred embodiments of the present application have been described, those skilled in the art may make other 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 falling within the scope of the present application.

[0104] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for optimizing wind power prediction error control for black start of super capacity energy storage, characterized by: include, Collect multimodal data, perform multimodal data preprocessing, and input the preprocessed multimodal data into the multimodal variational autoencoder model for data fusion to generate a unified feature representation; The fused data are used to construct prediction models at short-term, medium-term and long-term time scales, and multi-scale time series analysis is performed in combination with wavelet transform. The constructed prediction model includes a short-term prediction model, a medium-term prediction model and a long-term prediction model. The short-term prediction model adopts the long short-term memory network LSTM, the medium-term prediction model adopts the autoregressive integrated moving average model ARIMA, and the long-term prediction model adopts the trend analysis method and combines the wavelet transform to perform multi-scale time series analysis as follows: Among them, Ψ(Y) is the prediction result after multi-scale time series analysis, S is the number of time scales, s is the sth time scale, T s is the time range of the sth time scale, N s is the number of data points in the sth time scale, W s is the weight matrix of the sth time scale, σ(·) is the activation function, γ s (·) is the information filtering function of the sth time scale, K s is the convolution kernel of the sth time scale, X is the input data, μ is is the mean of the ith data point in the sth time scale, Σ is is the covariance matrix of the ith data point in the sth time scale, φ s (·) is the trend analysis function of the s-th time scale, is the data point at the sth time scale, dt s is the integral variable of the sth time scale; the outer sum Indicates the sum of the prediction results of all time scales to ensure that the information of all time scales is taken into account; inner summation Indicates that all data points are summed at each time scale and the information of all data points is integrated; the weight matrix W s represents the weight at each time scale; the exponential function exp(·) represents the Gaussian distribution of the data and is used to add nonlinear transformations in the feature extraction process; the activation function Represents the operation process of a multi-layer convolutional neural network to extract information from the data, where L s is the number of convolutional layers, K sl is the convolution kernel of the lth layer at the sth time scale, X l-1 is the input feature of the l-1th layer, b sl is the bias of the lth layer at the sth time scale; Indicates trend analysis at each time scale, where α s and β s is the parameter of the linear trend; the specific expansion is: The construction of the prediction model includes analyzing and judging according to the prediction results, assuming that the time span is T, and the short-term, medium-term and long-term time windows are T short , T medium , T long ; Short-term volatility calculation: in, is the average of the forecast results in the short term; Medium-term trend slope calculation: in, is the average value of time in the medium-term time window, It is the average of the forecast results in the medium term; Long-term trend slope calculation: in, is the average value of time in the long-term time window, is the average of the forecast results in the long term; If the short-term volatility σ short Greater than the threshold σ threshold , and the frequency of fluctuations exceeds expectations, then the forecast results reflect short-term fluctuations; If β medium Greater than the threshold β medium-threshold , and σ medium Less than the threshold σ medium-threshold , then the forecast results reflect an obvious medium-term trend; If β long Greater than the threshold β long-threshold , and σ long Less than the threshold σ long-threshold , then the forecast results reflect the long-term trend; If σ short , β medium and β long Equilibrium, that is, σ short , β medium and β long If all are within the expected range, the prediction results reflect the balance of multi-scale information; Construct a twin network for similarity learning, input historical data sample pairs, output similarity scores, identify similar wind conditions in historical data, and optimize the prediction model; Establish a hierarchical Bayesian model, perform hierarchical processing and correction on the prediction error, use the Markov chain Monte Carlo method or variational inference method to solve the hierarchical Bayesian model, and obtain the corrected prediction results; The hierarchical Bayesian model includes a multi-level error distribution, and the hierarchical structure includes the overall error distribution, the error distribution in different time periods, and the error distribution under the marked wind conditions. The model is solved by using the Markov chain Monte Carlo method or the variational inference method. The Markov chain Monte Carlo method includes using random sampling technology to estimate the parameters of the Bayesian model, and optimizing the variational lower bound by the variational inference method to approximate the Bayesian model. The formula is as follows: in, is the corrected prediction result, Q is the number of data samples, T i is the time range of the i-th sample, y is the error vector, μ i is the mean of the i-th sample, Σ i is the covariance matrix of the i-th sample, k is the dimension of the error vector, G is the latent variable, θ is the parameter vector of the Bayesian model, P(G|θ) is the latent variable distribution under given parameters, and P(θ) is the prior distribution of the parameters; The error distribution is assumed to be Gaussian with mean μ as parameter i and the covariance matrix Σ i ; Gaussian distribution represents the Gaussian distribution of the error vector y; the integral of the Gaussian distribution represents the distribution characteristics of the error vector, and the integral of the latent variables and parameters represents the hierarchical structure and reasoning process of the Bayesian model; The hierarchical Bayesian model is modeled through the prior distribution and conditional distribution of the latent variable G and the parameter θ; Markov chain Monte Carlo MCMC is used to solve the hierarchical Bayesian model and estimate the parameters of the Bayesian model through random sampling technology.

2. The method for optimizing wind power prediction error control for black start of super capacity energy storage according to claim 1, characterized in that: The collected multimodal data includes meteorological data, historical wind power data and environmental sensor data; the meteorological data includes wind speed, wind direction, temperature and humidity; the historical wind power data includes power generation and wind speed records; the environmental sensor data includes temperature sensor and humidity sensor data; The multimodal data preprocessing includes data normalization, denoising and missing value completion, data format unification and data cleaning; The data normalization process converts data of different dimensions into the same dimension, the denoising process uses low-pass filters and mean filters, and the interpolation and filling methods are used to fill in missing values.

3. The method for optimizing wind power prediction error control for black start of super capacity energy storage according to claim 2, characterized in that: The multimodal variational autoencoder model includes an encoder, a decoder and a latent space; the encoder maps the multimodal data to the latent space, and the decoder restores the representation of the latent space to the original data; The generating of a unified feature representation includes constructing a multimodal variational autoencoder model data fusion formula as follows: The information filtering function γ is expressed as the operation process of a multi-layer convolutional neural network, specifically: Among them, Φ(Z) is the fused feature representation, N is the number of data modalities, i is the data index of the modality, X is the input data, and W i is the data weight matrix of the ith mode, σ(·) is the activation function, γ(·) is the complex information filtering function, μ ij is the mean of the jth data in the ith mode, σ ij is the standard deviation of the jth data in the ith mode, M is the data dimension of each mode; t is the time variable, which is used for integration operation; is the partial derivative with respect to time, dX is the integral variable of the data, P is the number of convolution kernels, and X k is the feature after convolution, M k is the mean vector, C k is the covariance matrix, L is the number of convolution layers; K l is the convolution kernel of the lth layer; X l-1 is the input feature of the l-1th layer; b l is the bias of the lth layer; * indicates the convolution operation; X j is the jth data point, representing the specific features of the multimodal input data.

4. The method for optimizing wind power prediction error control for black start of super capacity energy storage according to claim 3, characterized in that: The analysis and judgment based on the forecast results include: if the short-term volatility σ short Greater than the threshold σ threshold , and the frequency of fluctuations exceeds expectations, real-time data monitoring is continued, and the number of sensors and the frequency of data collection are increased; when frequency fluctuations beyond expectations are detected, the data collection frequency is increased to every minute or every second, and a rapid response mechanism is established, including automatic adjustment of system parameters and manual intervention; when the predicted fluctuation exceeds the preset threshold, the output power of the wind turbine and the charge and discharge rate of the super-capacity energy storage system are automatically adjusted, and the operation and maintenance personnel are notified to conduct on-site inspections; if short-term fluctuations continue to intensify, the short-term prediction model LSTM is optimized, and the responsiveness of the model is improved by introducing new real-time data and repeated training; when the optimization is completed, the new model is deployed to the real-time monitoring system, and the model performance is regularly evaluated and adjusted; if the short-term fluctuation still exceeds the threshold, the alarm system is triggered, and a multi-level alarm mechanism is set to notify operation and maintenance personnel at different levels according to the degree of fluctuation; when the alarm system is started, the corresponding data is automatically recorded, and a fluctuation analysis report is provided for reference by the operation and maintenance personnel; If the medium-term trend slope β medium Greater than the threshold β medium-threshold , and the standard deviation in the medium term is medium Less than the threshold σ medium-threshold , then establish a mid-term trend tracking system, including trend analysis module and report generation module; When a medium-term trend is detected, the trend changes are continuously monitored, and a medium-term trend analysis report is generated and updated weekly or monthly. The operating parameters of the medium-term trend tracking system are adjusted according to the trend changes, and the charging and discharging strategy of the energy storage system and the operating mode of the wind turbine are adjusted. When the adjustment is completed, the new parameters are applied for operation, and a regular parameter inspection and adjustment mechanism is set up. At the same time, the medium-term prediction model ARIMA is optimized to improve the prediction accuracy by adding historical data and improving the model structure. When the model optimization is completed, the model is verified and the performance is evaluated, and it is deployed to the corresponding equipment for real-time prediction, the medium-term operation strategy is optimized, and the maintenance plan and resource allocation are adjusted. When the strategy adjustment is completed, the new strategy is implemented. If the long-term trend slope β long Greater than the threshold β long-threshold , and the standard deviation in the long term σ long Less than the threshold σ long-threshold , then formulate long-term plans and strategies, including equipment updates, technology improvements and resource allocation, and adjust policies and operation strategies according to trends, and formulate new capacity planning and energy storage system configuration plans; when policy adjustments are completed, implement new capacity planning and energy storage system configuration plans, introduce historical data and improve model structures, regularly update long-term prediction models, conduct model verification and performance evaluation, and then deploy them to the corresponding energy storage system for long-term prediction; If the short-term volatility σ short , medium-term trend slope β medium and the long-term trend slope β long If there is no single scale dominating the balance, a comprehensive monitoring system is established, including short-term, medium-term and long-term monitoring modules; when the balance information is detected, the short-term, medium-term and long-term changes are monitored simultaneously, a comprehensive analysis report is generated, and the short-term, medium-term and long-term prediction models are optimized by introducing multi-scale data and improving the model structure; when the model optimization is completed, the model is verified and the performance is evaluated, and then deployed to the comprehensive monitoring system for comprehensive prediction and the formulation of collaborative strategies; When the strategy is formulated, the coordinated strategy is implemented, and the system operating parameters are flexibly adjusted according to the multi-scale information, and the charging and discharging strategy of the energy storage system and the operating mode of the wind turbine are adjusted; when the adjustment is completed, the optimal operation plan is implemented.

5. The method for optimizing wind power prediction error control for black start of super capacity energy storage according to claim 4, characterized in that: The construction of the twin network for similarity learning includes that the twin network is composed of two identical neural networks, two different samples are input, and similarity is output. The twin network is trained by contrastive learning, and positive sample pairs and negative sample pairs are input to minimize the distance of the positive sample pairs and maximize the distance of the negative sample pairs: in, is the loss function, represents a positive sample pair, represents a negative sample pair, T is the time range, N1 is the number of data points, f(x) is the feature vector of sample x, S i is the similarity score of the i-th data point; The Siamese network is optimized by calculating the loss function, which is composed of the sum of the losses of the positive sample pair and the negative sample pair; the loss of the positive sample pair is achieved by minimizing the distance between two similar samples. Sum, calculate the time integral and nested sum of the feature vector difference; the loss of negative sample pairs is achieved by maximizing the distance between two dissimilar samples, by Perform summation, compute time integrals of eigenvector differences, and nested summations; The core of the twin network is to calculate the feature vector f(x) of the input sample. The difference of the feature vector is calculated through the partial derivative of time to reflect the change trend between samples. Activation Function The information filtering function is Where α is the filter parameter and a is the input; The final expansion formula is: Output is the value of the loss function, and its range is the non-negative real number set [0,∞); the smaller the value, the higher the similarity between the positive sample pairs, the greater the difference between the negative sample pairs, and the better the network recognition effect; the larger the value, the worse the network recognition effect.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of a method for optimizing the control of wind power prediction errors for black start of ultra-capacity energy storage are implemented as described in any one of claims 1 to 5.

7. 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 a method for optimizing the control of wind power prediction errors for black start of ultra-capacity energy storage are implemented as described in any one of claims 1 to 5.

Citation Information

Patent Citations

  • Ultra-short-term wind power plant power multi-step interval prediction modularization method and equipment thereof

    CN116070769A

  • Scheduling method and device based on wind power uncertainty, medium and electronic equipment

    CN117039896A

  • Black-start wind power prediction method based on super-capacity energy storage

    CN118523318A

  • Distributed industrial energy operation optimization platform automatically constructing intelligent models and algorithms

    US11487273B1

  • Method and system of predicting electric system load based on wavelet noise reduction and EMD-arima

    US20220045509A1