Load demand prediction method and system for distributed new energy power generation system

Abstract

The invention relates to the technical field of data processing, and discloses a load demand prediction method and system for a distributed new energy power generation system, and the method comprises the steps: obtaining historical data corresponding to power generation equipment and a load cluster in the distributed new energy power generation system, extracting feature parameters from the historical data, and carrying out the clustering analysis, dividing the power generation equipment and the load cluster into different subsets; processing the historical data corresponding to the power generation subsets and the load subsets to obtain a power generation output curve and a power utilization load curve; by analyzing the relevance between historical power generation and consumption data, obtaining a relevance index so as to fuse a power generation output curve and a power consumption load curve into a power generation load fusion model; and a parameter identification method is used to optimize the power abrupt change rate and the load change rate corresponding to the historical power generation data and the historical power utilization data, so that the power generation load fusion model is integrated into a generalized load model structure, a fusion equivalent model is obtained, and the load demand of the distributed new energy power generation system is accurately predicted.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a load demand forecasting method and system for a distributed new energy power generation system. Background Art

[0002] With the continuous development of power generation technology, the electricity demand of load clusters in distributed renewable energy power generation systems shows time-varying characteristics such as periodicity and trend. In order to rationally plan the power generation plan of distributed renewable energy power generation systems, it is necessary to accurately predict their load demand.

[0003] At present, time series analysis methods are usually used, such as ARIMA model, Fourier series modeling, etc., to predict the load demand of distributed renewable energy power generation systems. However, it is difficult to adapt to the rapid changes and uncertainties in the load demand of distributed renewable energy power generation systems. At the same time, this method has a strong dependence on historical data, resulting in inaccurate prediction results.

[0004] It can be seen that how to improve the prediction accuracy of load demand for distributed renewable energy power generation systems has become a technical problem that needs to be urgently solved by those skilled in the art. Summary of the invention

[0005] The present invention provides a method and system for predicting load demand of a distributed renewable energy power generation system, which solves the problem of how to improve the prediction accuracy of load demand of a distributed renewable energy power generation system.

[0006] In order to solve the above technical problems, the first aspect of the present invention provides a load demand forecasting method for a distributed renewable energy power generation system, comprising: Acquire historical power generation data of power generation equipment in a distributed renewable energy power generation system and historical power consumption data of a load cluster, and extract characteristic parameters from the historical power generation data and the historical power consumption data respectively to construct a characteristic parameter set of the distributed renewable energy power generation system; Performing cluster analysis on the characteristic parameter set, and dividing the power generation equipment and the load cluster into a plurality of power generation equipment cluster subsets and a plurality of load cluster cluster subsets respectively according to the cluster analysis result; Processing the specific historical power generation data corresponding to each of the power generation equipment cluster subsets to obtain the power generation output curve of each of the power generation equipment cluster subsets, and processing the specific historical power consumption data corresponding to each of the load cluster cluster subsets to obtain the power consumption load curve of each of the load cluster cluster subsets; Analyze the correlation between the historical power generation data and the historical power consumption data to obtain a correlation index reflecting the relationship between the power generation output and load fluctuation of the distributed new energy power generation system, and fuse each of the power generation output curves and each of the power consumption load curves according to the correlation index to obtain a power generation load fusion model; The power mutation rate and load change rate corresponding to the historical power generation data and the historical power consumption data are optimized through the parameter identification method, and based on the optimization result, the power generation load fusion model is integrated into the preset generalized load model structure to obtain the fusion equivalent model to predict the load demand of the distributed new energy power generation system.

[0007] As one of the preferred solutions, the extracting characteristic parameters from the historical power generation data and the historical power consumption data to construct a characteristic parameter set of the distributed new energy power generation system includes: Normalizing and inspecting the historical power generation data and the historical power consumption data respectively to obtain processed historical power generation data and processed historical power consumption data; The power fluctuation, power change rate, intermittent characteristic value and randomness index of the processed historical power generation data are calculated respectively as power generation characteristic parameters; Extracting the periodic variation characteristic value of the processed historical power consumption data based on discrete Fourier transform, and classifying the processed historical power consumption data according to the load type, so as to extract the peak and valley position of the historical power consumption curve corresponding to each type of load, and taking the periodic variation characteristic value as the load characteristic parameter; The radial basis function neural network is trained with the power generation characteristic parameters as input and the load characteristic parameters as output to obtain a characteristic parameter set; the characteristic parameter set is configured as an associated expression of the power generation characteristics and the power consumption characteristics in the distributed new energy power generation system.

[0008] As one of the preferred solutions, cluster analysis is performed on the characteristic parameter set, and the power generation equipment and the load cluster are divided into a plurality of power generation equipment cluster subsets and a plurality of load cluster cluster subsets according to the cluster analysis result, including: The characteristic parameter set is smoothed by a Butterworth low-pass filter, and the hourly mean value, daily power standard deviation and hourly power change rate of the power data in the processed characteristic parameter set are extracted by a sliding window to construct a power generation index sequence; Based on the probability distribution characteristics of the power generation index sequence according to quantile statistics, the Euclidean distance between each of the power generation equipment is calculated through hierarchical clustering according to the probability distribution characteristics to compare with the distance threshold, and each of the power generation equipment is divided according to the comparison result to obtain a plurality of power generation equipment cluster subsets.

[0009] As one of the preferred solutions, cluster analysis is performed on the characteristic parameter set, and the power generation equipment and the load cluster are divided into a plurality of power generation equipment cluster subsets and a plurality of load cluster cluster subsets according to the cluster analysis result, and further includes: Performing wavelet decomposition on the characteristic parameter set to obtain the load peak duration, peak-to-valley difference value and load root mean square error value of each load cluster at different time scales to construct a load characteristic vector; A self-organizing map network is used to cluster the load feature vector, the number of nodes in the competition layer of the self-organizing map network is used as the preset clustering number, and the similarity between the input vector and the weight vector of the self-organizing map network is calculated by the Euclidean distance between each load cluster, so as to update the weights of the winning node and its neighboring nodes in the self-organizing map network, and obtain several load cluster clustering subsets.

[0010] As one of the preferred solutions, the specific historical power generation data corresponding to each of the power generation equipment cluster subsets is processed to obtain the power generation output curve of each of the power generation equipment cluster subsets, including: The historical power generation data of each power generation device contained in each of the power generation device cluster subsets is used as the specific historical power generation data of the corresponding power generation device cluster subset, and each of the specific historical power generation data is normalized to obtain a standard power time series data set of each of the power generation device cluster subsets; Performing a three-layer decomposition on each of the standard power time series data sets by using a wavelet basis function to extract the power fluctuation characteristics of each of the power generation equipment cluster subsets, and extracting the principal components in the power fluctuation characteristics based on a singular spectrum analysis method to obtain the power fluctuation characteristic value of each power generation equipment in each of the power generation equipment cluster subsets; According to each of the power fluctuation characteristic values ​​and their corresponding normal power generation condition Gaussian distribution components and intermittent power generation condition Gaussian distribution components, a mixed Gaussian function is constructed, and the distribution parameters of the mixed Gaussian function are solved based on the maximization of the log-likelihood function, so as to describe the transition law of the power generation equipment between multiple power generation states through a Markov chain, and obtain the random power generation characteristic parameters of each of the cluster subsets of the power generation equipment; Quantifying the power mean of the power generation equipment contained in each of the power generation equipment cluster subsets, and quantifying the root mean square error between the specific historical power generation data of a single power generation equipment in the same power generation equipment cluster subset and the power mean, so as to use the power generation equipment with the smallest root mean square error as the representative power generation equipment of the corresponding power generation equipment cluster subset; The radial basis kernel function is used to interpolate and fit the specific historical power generation data and corresponding random power generation characteristic parameters of the representative power generation equipment in each of the power generation equipment cluster subsets to obtain the power generation output curve of each of the power generation equipment cluster subsets.

[0011] As one of the preferred solutions, the specific historical power consumption data corresponding to each of the load cluster subsets is processed to obtain the power load curve of each of the load cluster subsets, including: The historical power consumption data of each load cluster contained in each of the load cluster cluster subsets is used as the specific historical power consumption data of the corresponding load cluster cluster subset, and each of the specific historical power consumption data is season-marked, workday attribute-marked, and normalized to obtain a standard load time series data set of each of the load cluster cluster subsets; Performing three-layer wavelet decomposition on each of the standard load time series data sets to obtain load fluctuation characteristics on multiple time scales, and performing peak detection on each of the standard load time series data sets to obtain the start and end positions and duration of the load peak period, so as to construct the load shape feature vector of each of the load clustering subsets with the load fluctuation characteristics; The support vector regression of the radial basis kernel function is used to calculate the similarity between the characteristic vectors of each load form as a similarity index, and the correlation index between each load cluster is quantified based on the overlap of the start and end positions and duration of each load peak period, so as to construct a diversity feature set of each standard load time series data set with the similarity index; Based on each of the diversity feature sets, a recursive neural network with a double-layer cyclic structure is used to establish the power load curve of each of the load cluster cluster subsets; the first cyclic hidden layer of the recursive neural network is used to extract the load morphological characteristics of each of the load cluster cluster subsets, and the second cyclic hidden layer of the recursive neural network is used to extract the time-varying law of power consumption of each of the load cluster cluster subsets.

[0012] As one of the preferred solutions, the correlation between the historical power generation data and the historical power consumption data is analyzed to obtain a correlation index reflecting the relationship between the power generation output and load fluctuation of the distributed new energy power generation system, including: The historical power generation data and the historical power consumption data are processed respectively to obtain a historical power generation sequence and a historical load characteristic sequence, and the historical power generation sequence and the historical load characteristic sequence are time-aligned using a time sliding window to calculate the Pearson correlation coefficient in each sliding window, and a fluctuation synchronization curve for characterizing the correlation between power generation and load changes is constructed based on the correlation coefficient calculation result; Quantify the fluctuation trend characteristic value of the fluctuation synchronization curve within a preset time interval, and perform multi-scale analysis on the historical power generation sequence and the historical load characteristic sequence through three-layer wavelet decomposition to obtain a multi-scale sequence entropy value; Based on the multi-scale sequence entropy value and the fluctuation trend characteristic value, a correlation characteristic vector is constructed, and the correlation characteristic vector is used as a correlation index reflecting the power generation output and load fluctuation of the distributed new energy power generation system.

[0013] As one of the preferred solutions, the generation output curves and the power load curves are fused according to the associated index to obtain a generation load fusion model, including: Performing joint dimension reduction and restoration reconstruction processing on each of the power output curves and each of the power load curves through a codec to obtain a power generation load fusion feature to construct a feature association matrix with the association index; Performing singular value decomposition on the feature association matrix to obtain a feature subspace, and selecting basis vectors corresponding to feature association values ​​whose cumulative contribution rates reach a preset contribution threshold from the feature subspace as matching features, so as to calculate the matching degree between each of the power generation equipment cluster subsets and each of the load cluster cluster subsets; The power generation output curves of the power generation equipment cluster subset whose matching degree exceeds a preset matching threshold are fused with the power consumption load curves of the load cluster cluster subset to obtain the power generation load fusion model.

[0014] As one of the preferred solutions, the power mutation rate and load change rate corresponding to the historical power generation data and the historical power consumption data are optimized by a parameter identification method, and the power generation load fusion model is integrated into a preset generalized load model structure based on the optimization result to obtain a fusion equivalent model, including: Extracting historical power mutation rate and historical load change rate from the historical power generation data and the historical power consumption data respectively, and performing sensitivity calculation on the historical power mutation rate and the historical load change rate by using a cross-validation method to obtain a parameter sensitivity index; Taking the parameter sensitivity index as a weight coefficient, optimizing the historical power mutation rate by a recursive least square method with the weight coefficient, and optimizing the historical load change rate according to the parameter sensitivity index; Support vector regression fitting is performed on the optimized historical power mutation rate and the optimized historical load change rate, and a quadratic generalized load expression including a power response term and a load response term is constructed according to the fitting results to obtain the fusion equivalent model.

[0015] A second aspect of the present invention provides a load demand forecasting system for a distributed renewable energy power generation system, comprising: A parameter set building module, used to obtain historical power generation data of power generation equipment in a distributed renewable energy power generation system and historical power consumption data of a load cluster, and extract characteristic parameters from the historical power generation data and the historical power consumption data to build a characteristic parameter set of the distributed renewable energy power generation system; A cluster subset division module, used for performing cluster analysis on the characteristic parameter set, and dividing the power generation equipment and the load cluster into a plurality of power generation equipment cluster subsets and a plurality of load cluster cluster subsets respectively according to the cluster analysis result; A subset curve construction module is used to process the specific historical power generation data corresponding to each of the power generation equipment cluster subsets to obtain the power generation output curve of each of the power generation equipment cluster subsets, and to process the specific historical power consumption data corresponding to each of the load cluster cluster subsets to obtain the power consumption load curve of each of the load cluster cluster subsets; A fusion model generation module is used to analyze the correlation between the historical power generation data and the historical power consumption data to obtain a correlation index reflecting the relationship between the power generation output and load fluctuation of the distributed new energy power generation system, and fuse each of the power generation output curves and each of the power consumption load curves according to the correlation index to obtain a power generation load fusion model; The load demand prediction module is used to optimize the power mutation rate and load change rate corresponding to the historical power generation data and the historical power consumption data respectively through the parameter identification method, and integrate the power generation load fusion model into the preset generalized load model structure based on the optimization result to obtain the fusion equivalent model to predict the load demand of the distributed new energy power generation system.

[0016] Compared with the prior art, the embodiments of the present invention have the following advantages: (1) By obtaining the historical data corresponding to the power generation equipment and load clusters in the distributed renewable energy power generation system, it is possible to perform analysis based on the actual operation data, thereby improving the accuracy of the prediction model; extracting characteristic parameters from the historical data and dividing the power generation equipment and load clusters into different subsets through cluster analysis helps to identify equipment or clusters with similar characteristics, thereby simplifying the complexity of subsequent analysis and improving processing efficiency.

[0017] (2) The historical data of each power generation equipment cluster subset and load cluster subset are processed to obtain clear power generation output curve and power load curve, which makes the power generation and power consumption behavior of the system clearer and helps to better understand the operating laws of the system; by analyzing the correlation between power generation data and power consumption data, the correlation index reflecting the system power generation output and load fluctuation is obtained, which helps to reveal the dynamic balance relationship within the system and provides a scientific basis for the fusion of models.

[0018] (3) The parameter identification method is used to optimize the power mutation rate and load change rate corresponding to the historical data, which can further improve the accuracy and adaptability of the model. Based on the optimization results, the power generation load fusion model is integrated into the generalized load model structure to obtain a fusion equivalent model that can comprehensively consider the changes in renewable energy power generation and load demand. By constructing a comprehensive model that can reflect the randomness of renewable energy output and capture the time-varying nature of load, the distributed renewable energy power generation system can be accurately simulated to predict its load demand, thereby improving the reliability and accuracy of the prediction and providing a strong guarantee for the economic, safe and stable operation of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solution of the present invention, the drawings required for use in the implementation mode will be briefly introduced below. Obviously, the drawings described below are only some implementation modes of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 It is a flow chart of a load demand prediction method for a distributed renewable energy power generation system provided by a certain embodiment of the present invention; Figure 2 It is a structural diagram of a load demand prediction system for a distributed renewable energy power generation system provided by a certain embodiment of the present invention. DETAILED DESCRIPTION

[0021] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings and embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] In the description of this application, the terms "first", "second", "third", etc. are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first", "second", "third", etc. may explicitly or implicitly include one or more of the feature. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0023] In the description of the present application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two components. The terms "vertical", "horizontal", "left", "right", "upper", "lower" and similar expressions used herein are for illustrative purposes only, and do not indicate or imply that the system or component referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0024] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those commonly understood by those skilled in the art. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood by specific circumstances.

[0025] In one embodiment, if Figure 1 As shown, the first aspect of the present invention provides a load demand forecasting method for a distributed renewable energy power generation system, comprising: S1. Obtain the historical power generation data of the power generation equipment in the distributed renewable energy power generation system and the historical power consumption data of the load cluster, and extract characteristic parameters from the historical power generation data and the historical power consumption data respectively to construct a characteristic parameter set of the distributed renewable energy power generation system; wherein, the power generation equipment in the distributed renewable energy power generation system may be a photovoltaic power station, a wind turbine generator set, etc., and the corresponding historical power generation data is the historical power output data of these power generation equipment, which can also be regarded as historical operation data; the types of load clusters in the distributed renewable energy power generation system include industrial loads, commercial loads, and residential loads, etc., and the corresponding historical power consumption data is the historical power consumption data.

[0026] In one embodiment, extracting characteristic parameters from the historical power generation data and the historical power consumption data to construct a characteristic parameter set of the distributed new energy power generation system includes: Normalizing and inspecting the historical power generation data and the historical power consumption data respectively to obtain processed historical power generation data and processed historical power consumption data; The power fluctuation, power change rate, intermittent characteristic value and randomness index of the processed historical power generation data are calculated respectively as power generation characteristic parameters; Extracting the periodic variation characteristic value of the processed historical power consumption data based on discrete Fourier transform, and classifying the processed historical power consumption data according to the load type, so as to extract the peak and valley position of the historical power consumption curve corresponding to each type of load, and taking the periodic variation characteristic value as the load characteristic parameter; The radial basis function neural network is trained with the power generation characteristic parameters as input and the load characteristic parameters as output to obtain a characteristic parameter set; the characteristic parameter set is configured as an associated expression of the power generation characteristics and the power consumption characteristics in the distributed new energy power generation system.

[0027] Specifically, after acquiring the historical power consumption data, the present invention also needs to identify the load type, and then normalize the collected historical data to remove the dimension effect, and eliminate abnormal values ​​through data quality inspection, and store the data in association according to the running timestamp. Then, the standard deviation of the processed historical power consumption data is calculated in the sliding time window to obtain the power fluctuation degree to reflect the degree of fluctuation of the power generation equipment under different weather conditions; the processed historical power consumption data is differentially calculated to obtain the power change rate; the intermittent characteristic value of the power generation equipment is obtained by performing amplitude statistics on the processed historical power consumption data; the randomness index of the power generation equipment is calculated based on the deviation between the power output prediction value and the actual value of the processed historical power consumption data to reflect the unpredictability or uncertainty of the power generation equipment when outputting electricity; and the power fluctuation degree, power change rate, intermittent characteristic value and randomness index are used as power generation characteristic parameters, that is, the power generation power and output fluctuations in the historical power generation data of the power generation equipment reflect the randomness and intermittency of the new energy power generation equipment.

[0028] Since the power output fluctuation of power generation equipment in distributed renewable energy power generation systems comes from changes in the meteorological environment, taking wind turbines as an example, in a wind farm, when the wind speed drops from 4 meters per second to 2 meters per second, the wind turbine power output drops from 30% of the rated power to 10%, showing the intermittent characteristics of the power generation equipment. The power sampling data at 30-second intervals is used for differential calculation, and the power change rate is -40% per minute, marking this type of power change as an intermittent event; the power output prediction value can be predicted by regression analysis, time series analysis, gray prediction method, machine learning algorithm, etc., or by a power prediction algorithm based on weather forecast, with a prediction lead time of 24 hours. In actual operation, the power output of the photovoltaic power station deviates from the predicted value, with a mean deviation of 15% and a standard deviation of 0.2. A randomness index is constructed based on the deviation statistics to characterize the random characteristics of photovoltaic power generation output.

[0029] Then, time series clustering based on dynamic time warping distance is used to classify the historical electricity consumption data of the load cluster according to the load type identification, and the peak and valley identification is performed on the electricity load curve formed by the historical electricity consumption data of each type of load to mark the peak period. The periodic change characteristic values ​​of the historical electricity consumption data of the load cluster are extracted based on discrete Fourier transform to reflect the differences in electricity consumption patterns of different types of loads. The peak and valley positions of the electricity load curve corresponding to each type of load cluster and its periodic change characteristic values ​​are used as the load characteristic parameters of the corresponding load cluster, that is, the diversity and time variability of the load cluster are reflected by the load curve shape and peak electricity consumption period.

[0030] Finally, the power generation characteristic parameters are used as input features and the load characteristic parameters are used as output features. The radial basis function neural network with 1 hidden layer is used for training. The Gaussian function is used as the radial basis function. Different variance parameters are set in the hidden layer nodes. The variance values ​​range from 0.1 to 0.9. A total of 8 hidden layer nodes are set to fit the mapping relationship between the power generation fluctuation characteristics and the load curve characteristics. When a new power generation sequence is input, the neural network can output the corresponding load characteristic parameters, realize the associated expression of the power generation characteristics and power consumption characteristics of the distributed new energy power generation system, and obtain the characteristic parameter set of the power generation equipment and the load cluster. The present invention improves the comparability and analysis efficiency of data by preprocessing and extracting features from historical data; constructs the correlation expression between power generation characteristics and power consumption characteristics through the training of RBF neural network, which helps to better understand the operation law and demand changes of the power system; the constructed characteristic parameter set can provide a scientific basis and decision-making support for the planning, operation and control of the power system, which helps to improve the stability, reliability and economy of the power system.

[0031] S2. performing cluster analysis on the characteristic parameter set, and dividing the power generation equipment and the load cluster into a plurality of power generation equipment cluster subsets and a plurality of load cluster cluster subsets respectively according to the cluster analysis result; In one embodiment, cluster analysis is performed on the characteristic parameter set, and the power generation equipment and the load cluster are divided into a plurality of power generation equipment cluster subsets and a plurality of load cluster cluster subsets according to the cluster analysis result, including: The characteristic parameter set is smoothed by a Butterworth low-pass filter, and the hourly mean value, daily power standard deviation and hourly power change rate of the power data in the processed characteristic parameter set are extracted by a sliding window to construct a power generation index sequence; Based on the probability distribution characteristics of the power generation index sequence according to quantile statistics, the Euclidean distance between each of the power generation equipment is calculated through hierarchical clustering according to the probability distribution characteristics to compare with the distance threshold, and each of the power generation equipment is divided according to the comparison result to obtain a plurality of power generation equipment cluster subsets.

[0032] Specifically, the present invention smoothes the characteristic parameter set through a Butterworth low-pass filter with a cutoff frequency set to 0.1 Hz to filter out high-frequency disturbance components and retain the main change trend of the generated power, thereby extracting the time series power data of the power generation equipment, and using a sliding window to calculate the hourly mean value, daily power standard deviation, and hourly power change rate of the time series power data, and uses it as a power generation indicator sequence that characterizes the randomness and intermittent characteristics of the power generation equipment.

[0033] For each power generation index sequence, the probability distribution characteristics including mean point, fluctuation amplitude point and change frequency point are calculated based on quantile statistics. Then, based on the probability distribution characteristics, the Euclidean distance between each power generation equipment is calculated through hierarchical clustering to establish a Euclidean distance matrix, and the data in the distance matrix is ​​compared with the preset distance threshold. The cluster center point corresponding to the distance value less than the preset distance threshold is selected as the basis for division to divide each power generation equipment, and several cluster subsets of power generation equipment with similar power generation characteristics are obtained. Among them, the distance matrix between power generation equipment is calculated in the hierarchical clustering process. The distance value of adjacent power generation equipment is 0.2, and the distance value of dissimilar power generation equipment exceeds 0.8. When the distance threshold is set to 0.4, taking photovoltaic power generation as an example, it can be divided into 3 cluster subsets, corresponding to large ground power stations, distributed rooftop power stations and agricultural photovoltaic complementary power stations, to reflect the differences in power generation characteristics of different types of photovoltaic power stations.

[0034] The present invention smoothes the characteristic parameter set through a Butterworth low-pass filter, effectively removing high-frequency noise and spike interference in the data, and improving the smoothness and analyzability of the data; uses a sliding window to extract the power data in the processed characteristic parameter set, constructs a power generation index sequence, and provides a data basis for the subsequent division and optimization of power generation equipment; divides the power generation equipment based on quantile statistics and hierarchical clustering, and obtains several cluster subsets of power generation equipment, which is helpful to identify power generation equipment with similar operating characteristics and improve management efficiency and power generation benefits.

[0035] In one embodiment, the cluster analysis is performed on the characteristic parameter set, and the power generation equipment and the load cluster are divided into a plurality of power generation equipment cluster subsets and a plurality of load cluster cluster subsets according to the cluster analysis result, and further includes: Performing wavelet decomposition on the characteristic parameter set to obtain the load peak duration, peak-to-valley difference value and load root mean square error value of each load cluster at different time scales to construct a load characteristic vector; A self-organizing map network is used to cluster the load feature vector, the number of nodes in the competition layer of the self-organizing map network is used as the preset clustering number, and the similarity between the input vector and the weight vector of the self-organizing map network is calculated by the Euclidean distance between each load cluster, so as to update the weights of the winning node and its neighboring nodes in the self-organizing map network, and obtain several load cluster clustering subsets.

[0036] Specifically, after preprocessing the characteristic parameter set, the present invention extracts the load time series data therein for wavelet decomposition. After wavelet decomposition, the load time series data obtains component values ​​of four scales. The first-level components reflect the short-term fluctuations of the load, the second-level components reflect the intraday variation law, the third-level components characterize the periodic variation characteristics, and the fourth-level components correspond to seasonal variations. The component values ​​of the load curve on different time scales can be obtained, that is, the load peak duration and peak-to-valley difference are extracted on the hourly scale, and the root mean square error of the load on working days and rest days is statistically calculated on the daily scale, and constructed as a load characteristic vector.

[0037] Then, a self-organizing map network is used to cluster the load feature vector. The number of nodes in the input layer of the network is the same as the dimension of the feature vector, and the number of nodes in the competition layer is the preset number of clusters. The similarity between the input vector and the weight vector is calculated by Euclidean distance, and the weights of the winning node and its neighboring nodes are updated to obtain several load clustering subsets with similar load characteristics. Among them, the input layer of the self-organizing map network contains 8 nodes, corresponding to the peak-to-valley difference, peak duration, daily average load and other characteristic quantities of the load curve. The competition layer sets 16 nodes to form a 4x4 two-dimensional plane. The topological distance between adjacent nodes is 1. The weight adjustment range of the winning node in training is 0.1, and the weight adjustment range of the neighboring node gradually decreases with the increase of distance. The manufacturing load in the industrial park can be divided into the same clustering subset, reflecting similar electricity consumption characteristics. The office buildings and shopping malls in the commercial area are classified into another clustering subset. The residential community is divided into two clustering subsets, large communities and communities, according to the scale of electricity consumption. The clustering results reflect the differences in electricity consumption characteristics of different types of loads. When new load curve data is input, the similarity between the load and each cluster center is calculated. The similarity indicators include power correlation coefficient, peak-valley characteristic similarity, load duration rate, etc. The cluster subset with the highest similarity is the category to which the load belongs, realizing automatic recognition and classification of load characteristics.

[0038] The present invention constructs load characteristic vectors through wavelet decomposition, which can more accurately extract key characteristic parameters of the load and provide strong data support for subsequent cluster analysis; adopts self-organizing map network for cluster analysis, which can adaptively determine the number of clusters and avoid subjectivity caused by human settings; the division of load cluster cluster subsets can reflect the characteristics and requirements of different load clusters, and provide strong support for load management and optimization decisions.

[0039] S3, processing the specific historical power generation data corresponding to each of the power generation equipment cluster subsets to obtain the power generation output curve of each of the power generation equipment cluster subsets, and processing the specific historical power consumption data corresponding to each of the load cluster cluster subsets to obtain the power consumption load curve of each of the load cluster cluster subsets; In one embodiment, the processing of the specific historical power generation data corresponding to each of the power generation equipment cluster subsets to obtain the power generation output curve of each of the power generation equipment cluster subsets includes: The historical power generation data of each power generation device contained in each of the power generation device cluster subsets is used as the specific historical power generation data of the corresponding power generation device cluster subset, and each of the specific historical power generation data is normalized to obtain a standard power time series data set of each of the power generation device cluster subsets; Performing a three-layer decomposition on each of the standard power time series data sets by using a wavelet basis function to extract the power fluctuation characteristics of each of the power generation equipment cluster subsets, and extracting the principal components in the power fluctuation characteristics based on a singular spectrum analysis method to obtain the power fluctuation characteristic value of each power generation equipment in each of the power generation equipment cluster subsets; According to each of the power fluctuation characteristic values ​​and their corresponding normal power generation condition Gaussian distribution components and intermittent power generation condition Gaussian distribution components, a mixed Gaussian function is constructed, and the distribution parameters of the mixed Gaussian function are solved based on the maximization of the log-likelihood function, so as to describe the transition law of the power generation equipment between multiple power generation states through a Markov chain, and obtain the random power generation characteristic parameters of each of the cluster subsets of the power generation equipment; Quantifying the power mean of the power generation equipment contained in each of the power generation equipment cluster subsets, and quantifying the root mean square error between the specific historical power generation data of a single power generation equipment in the same power generation equipment cluster subset and the power mean, so as to use the power generation equipment with the smallest root mean square error as the representative power generation equipment of the corresponding power generation equipment cluster subset; The radial basis kernel function is used to interpolate and fit the specific historical power generation data and corresponding random power generation characteristic parameters of the representative power generation equipment in each of the power generation equipment cluster subsets to obtain the power generation output curve of each of the power generation equipment cluster subsets.

[0040] Specifically, the present invention uses the historical power generation data of each power generation device contained in each power generation device cluster subset as the specific historical power generation data of the corresponding power generation device cluster subset, and aligns each specific historical power generation data according to the timestamp through a sliding window. If the data missing time exceeds 3600 seconds, the data of this period is discarded. If the data value deviates from the mean by more than 3 times the standard deviation, it is marked as an outlier and eliminated. After linear normalization processing, a standard power time series data set is obtained.

[0041] The db4 wavelet basis function is used to perform a three-layer decomposition of the standard power time series data set to extract the power fluctuation characteristics of each cluster subset of power generation equipment; among them, taking photovoltaic power generation as an example, in the wavelet decomposition process of photovoltaic power generation power data, the first layer coefficient reflects the short-term fluctuation caused by cloud cover, and the fluctuation period is within 5 minutes, the second layer coefficient reflects the medium-term fluctuation caused by the change of the solar altitude angle, and the fluctuation period is about 1 hour, and the third layer coefficient corresponds to the long-term fluctuation caused by weather changes, and the fluctuation period is more than 4 hours. Taking wind power generation as an example, its first layer coefficient corresponds to the power fluctuation caused by wind speed pulsation, the second layer coefficient reflects the impact of local weather changes, and the third layer coefficient reflects the evolution characteristics of the weather system. Then, based on singular spectrum analysis, the three principal components with the largest contribution rate in the power fluctuation characteristics are extracted. The first principal component reflects the sunshine law, the second principal component reflects the weather change, and the third principal component corresponds to the equipment characteristics, and the power fluctuation characteristic value of each power generation equipment in each cluster subset of power generation equipment is obtained.

[0042] A mixed Gaussian function is constructed according to the power fluctuation characteristic value and its corresponding Gaussian distribution component of normal power generation condition and intermittent power generation condition. The power distribution is relatively concentrated under normal condition with a standard deviation of 0.2, and more dispersed under intermittent condition with a standard deviation of 0.45. Then, the mixed weight value and distribution parameter of the mixed Gaussian function are solved by maximizing the log-likelihood function. The transfer law of the power generation between normal power generation, intermittent attenuation, complete intermittent and restored power generation states is described based on the 4-state Markov chain. The 4 states in the Markov chain characterize the operating characteristics of the power generation equipment, with the normal power generation state accounting for 70%, the intermittent attenuation state accounting for 15%, the complete intermittent state accounting for 10%, and the restored power generation state accounting for 5%. The random power generation characteristic parameters of the cluster subsets of each power generation equipment can be obtained.

[0043] Finally, the power data of all power generation equipment at each moment in each power generation equipment cluster subset are averaged, and the root mean square error between the power sequence of a single power generation equipment in the corresponding subset and the average power sequence is calculated. The power generation equipment with the smallest error value is selected as the representative power generation equipment of the subset, and the radial basis kernel function is used to interpolate and fit the power curve and random power generation characteristic parameters of the representative power generation equipment to obtain the typical power generation output curve of the corresponding cluster subset; among them, the bandwidth parameter of the radial basis kernel function is set to 0.3, and the interpolation node interval is set to 1800 seconds, which not only ensures the smoothness of the curve, but also reflects the dynamic characteristics of power changes.

[0044] The present invention can improve the accuracy and efficiency of data analysis of power generation equipment by normalizing the specific historical power generation data of power generation equipment in the power generation equipment cluster subset; focus on any details of the data through wavelet basis function decomposition to accurately extract power fluctuation characteristics, and extract the main components in the power fluctuation characteristics based on the singular spectrum analysis method, which can further simplify the feature vector and improve the accuracy and efficiency of feature extraction; construct a mixed Gaussian function to describe the state of the power generation equipment, describe the transition law of the power generation equipment between multiple power generation states based on the Markov chain to reveal the dynamic process of the state change of the power generation equipment, and quantify the key information such as the transition probability and transition time of the power generation equipment between different states through the random power generation characteristic parameters obtained by the Markov chain, so as to provide an important basis for the state prediction and fault diagnosis of the power generation equipment; finally, based on the radial basis kernel function interpolation fitting to optimize the output curve of the power generation equipment, the equivalent model of each cluster subset of new energy power generation equipment reflecting the power generation of the cluster subset of new energy power generation equipment, the randomness and intermittency characteristics of the output fluctuation parameters is obtained.

[0045] In one embodiment, the processing of the specific historical power consumption data corresponding to each of the load cluster subsets to obtain the power consumption load curve of each of the load cluster subsets includes: The historical power consumption data of each load cluster contained in each of the load cluster cluster subsets is used as the specific historical power consumption data of the corresponding load cluster cluster subset, and each of the specific historical power consumption data is season-marked, workday attribute-marked, and normalized to obtain a standard load time series data set of each of the load cluster cluster subsets; Performing three-layer wavelet decomposition on each of the standard load time series data sets to obtain load fluctuation characteristics on multiple time scales, and performing peak detection on each of the standard load time series data sets to obtain the start and end positions and duration of the load peak period, so as to construct the load shape feature vector of each of the load clustering subsets with the load fluctuation characteristics; The support vector regression of the radial basis kernel function is used to calculate the similarity between the characteristic vectors of each load form as a similarity index, and the correlation index between each load cluster is quantified based on the overlap of the start and end positions and duration of each load peak period, so as to construct a diversity feature set of each standard load time series data set with the similarity index; Based on each of the diversity feature sets, a recursive neural network with a double-layer cyclic structure is used to establish the power load curve of each of the load cluster cluster subsets; the first cyclic hidden layer of the recursive neural network is used to extract the load morphological characteristics of each of the load cluster cluster subsets, and the second cyclic hidden layer of the recursive neural network is used to extract the time-varying law of power consumption of each of the load cluster cluster subsets.

[0046] Specifically, the present invention uses the historical electricity consumption data of each load cluster contained in each load cluster cluster subset as the specific historical electricity consumption data of the corresponding load cluster cluster subset, and performs data preprocessing on it, that is, supplementing the missing data points based on linear interpolation, using a median filter with a window length of 5 to remove outliers, marking the specific historical electricity consumption data with seasonal and working day attributes through 24-hour time period division, and obtaining a standard load time series data set based on maximum value normalization. Since the electricity load data is affected by multiple factors and presents different characteristics, the curve morphology of different types of loads is more obvious after data normalization.

[0047] Then, cubic spline decomposition is applied to the load curve corresponding to the standard time series data set to extract the seasonal component. The fluctuation characteristic values ​​of the load curve on the hourly, daily and weekly time scales are obtained through three-layer wavelet decomposition. The first layer reflects the start and stop characteristics of the load equipment, the second layer reflects the daily work and rest rules, and the third layer corresponds to the impact of weekends and holidays. The peak detection based on the 2 times standard deviation threshold is used to extract the start and end positions and duration of the peak period. The load change rate within the 24-hour interval is segmented and statistically analyzed to obtain the load shape characteristic quantity, so as to construct a load shape characteristic vector including the peak-to-valley difference, peak duration and load change rate.

[0048] Then, support vector regression with radial basis kernel function is used to calculate the similarity between the load shape feature vector and the cluster center vector. The bandwidth parameter is taken as 0.8, and the correlation index between loads is calculated based on the overlap of the start and end times during the peak period. The overlap of the same type of loads exceeds 0.7, and the overlap of different types of loads is below 0.3, which can reflect the diversity characteristics of the load curve shape. The diversity feature set of each load cluster cluster subset is constructed by combining the load similarity and correlation indicators.

[0049] Finally, a recursive neural network with a double-layer loop structure is used to establish a load curve equivalent model. The first layer of the loop hidden layer contains 128 nodes to extract load morphological characteristics, and the second layer of the loop hidden layer contains 64 nodes to extract the time-varying law of electricity consumption. The load distribution index is constructed according to the load mean, standard deviation, and peak-to-valley difference within a 24-hour interval, and the load equivalent model that characterizes the load diversity and time-varying characteristics is obtained, that is, the electricity load curve of each load cluster cluster subset; among them, the first layer of the 128 hidden layer nodes of the recursive neural network extracts load morphological characteristics, including peak-to-valley difference, peak duration, and load change rate, and the second layer of the 64 hidden layer nodes captures the load law, reflecting the influence of day and night alternation, work and rest changes, and seasonal changes.

[0050] The present invention realizes in-depth mining and effective utilization of historical power consumption data of load clusters through multiple steps such as precise data processing and feature extraction, construction of diverse feature sets, and establishment of power load curves using recursive neural networks. This can not only improve the accuracy and reliability of load forecasting, but also provide more refined decision support for the planning, scheduling and operation of power systems.

[0051] S4, analyzing the correlation between the historical power generation data and the historical power consumption data to obtain a correlation index reflecting the relationship between the power generation output and load fluctuation of the distributed new energy power generation system, and fusing each of the power generation output curves and each of the power consumption load curves according to the correlation index to obtain a power generation load fusion model; In one embodiment, the correlation between the historical power generation data and the historical power consumption data is analyzed to obtain a correlation index reflecting the relationship between the power generation output and load fluctuation of the distributed new energy power generation system, including: The historical power generation data and the historical power consumption data are processed respectively to obtain a historical power generation sequence and a historical load characteristic sequence, and the historical power generation sequence and the historical load characteristic sequence are time-aligned using a time sliding window to calculate the Pearson correlation coefficient in each sliding window, and a fluctuation synchronization curve for characterizing the correlation between power generation and load changes is constructed based on the correlation coefficient calculation result; Quantify the fluctuation trend characteristic value of the fluctuation synchronization curve within a preset time interval, and perform multi-scale analysis on the historical power generation sequence and the historical load characteristic sequence through three-layer wavelet decomposition to obtain a multi-scale sequence entropy value; Based on the multi-scale sequence entropy value and the fluctuation trend characteristic value, a correlation characteristic vector is constructed, and the correlation characteristic vector is used as a correlation index reflecting the power generation output and load fluctuation of the distributed new energy power generation system.

[0052] Specifically, the present invention reads the historical power generation sequence and the historical load characteristic sequence from the historical power generation data and the historical power consumption database at a sampling time interval of 5 minutes, removes the abnormal values ​​that deviate from the mean by more than 3 times the standard deviation, uses linear interpolation to supplement the missing data points, and calculates the average power of the new energy power generation equipment within the preset time according to the historical power generation sequence to obtain the power generation sequence. The historical load characteristic sequence within the preset time is subjected to maximum value normalization processing, and the load curve is detected by the local maximum value method. The data points that deviate from the mean by more than 2 times the standard deviation in the local interval are determined to be peak and valley points. The corresponding time and duration of the peak and valley points are statistically calculated, and the normalized characteristic value of the fluctuation amplitude is calculated based on the peak-valley difference to obtain the load fluctuation characteristic sequence that characterizes the intraday changes in the power grid load.

[0053] A sliding window with a length of 1 hour is used to time-align the power generation sequence and the load fluctuation characteristic sequence, so as to calculate the Pearson correlation coefficient in each sliding window, and the fluctuation synchronization curve of power generation and load changes is constructed according to the correlation coefficient sequence of 96 time slices, and the fluctuation trend characteristic value of the fluctuation synchronization curve in the 24-hour interval is calculated. The power generation sequence and the load fluctuation characteristic sequence are multi-scale analyzed by three-layer wavelet decomposition, and the fluctuation law is statistically analyzed on the three time scales of hour, day and day. Among them, the first layer coefficient reflects the rapid fluctuation within 15 minutes, the second layer coefficient corresponds to the change of 1 hour scale, and the third layer coefficient reflects the trend of 4 hours scale; the sequence entropy value is calculated for each time scale, and the multi-scale sequence entropy value is obtained to construct the correlation characteristic vector in combination with the fluctuation trend characteristic value of the fluctuation synchronization curve, so as to reflect the correlation index of power generation output and load fluctuation of distributed new energy power generation system, which integrates the time series correlation and fluctuation characteristics.

[0054] The present invention performs time alignment and Pearson correlation coefficient calculation on historical power generation sequences and historical load characteristic sequences through a time sliding window, which helps to identify the synchronization between power generation output and load fluctuations. The fluctuation synchronization curve constructed based on the correlation coefficient calculation result can intuitively display the correlation between power generation and load changes, which helps to discover the synchronous or asynchronous change pattern between the two; by quantitatively analyzing the fluctuation characteristics of power generation and load, it helps to identify potential risks and uncertainties in the power system; by constructing associated characteristic vectors, the system performance evaluation capability is improved, which helps to improve the overall performance and operation efficiency of the power system and promote the full utilization and sustainable development of new energy.

[0055] In one embodiment, the generation output curves and the power load curves are fused according to the associated index to obtain a generation load fusion model, including: Performing joint dimension reduction and restoration reconstruction processing on each of the power output curves and each of the power load curves through a codec to obtain a power generation load fusion feature to construct a feature association matrix with the association index; Performing singular value decomposition on the feature association matrix to obtain a feature subspace, and selecting basis vectors corresponding to feature association values ​​whose cumulative contribution rates reach a preset contribution threshold from the feature subspace as matching features, so as to calculate the matching degree between each of the power generation equipment cluster subsets and each of the load cluster cluster subsets; The power generation output curves of the power generation equipment cluster subset whose matching degree exceeds a preset matching threshold are fused with the power consumption load curves of the load cluster cluster subset to obtain the power generation load fusion model.

[0056] Specifically, the present invention adopts a deep autoencoder with 3 hidden layers and 256, 128 and 16 hidden layer nodes in the encoder to jointly reduce the dimension of each power output curve and each power load curve, and restores the 16-dimensional feature vector after dimension reduction through a decoder to obtain the power generation load fusion feature; wherein, in the 3-layer structure of the deep autoencoder, the 256 nodes in the first hidden layer extract the local features of the time series, the 128 nodes in the second hidden layer fuse the features of different time scales, and the 16 nodes in the third hidden layer realize feature dimension reduction; a feature correlation matrix is ​​constructed based on the power generation load fusion feature and the correlation index, and the correlation matrix is ​​subjected to singular value decomposition to extract the feature subspace The basis vectors corresponding to the eigenvalues ​​with cumulative contribution rates reaching 0.9 are selected as matching features to calculate the matching degree between each power generation equipment cluster subset and each load cluster cluster subset. If the matching degree exceeds 0.8, the power generation output curve of the corresponding power generation equipment cluster subset and the power load curve of the load cluster cluster subset are marked as matching objects and fused to obtain a power generation load fusion model, which includes the timing characteristics and matching characteristics of the power generation load. The timing characteristics describe the dynamic relationship between the power generation output and the load change, and the matching characteristics reflect the degree of adaptation between the power generation capacity and the electricity demand. The organic combination of the two types of features forms a complete description of the power generation load fusion.

[0057] The present invention uses a codec to jointly reduce the dimension and restore and reconstruct the power generation output curve and the power load curve, thereby improving the data processing efficiency and accuracy; the codec extracts the key features in the power generation output curve and the power load curve to form a power generation load fusion feature, so as to construct a feature association matrix with related indicators, which is helpful to explore the deep relationship between power generation and power consumption in the power generation system, and provides a basis for subsequent matching degree calculation and power generation load fusion model construction; singular value decomposition is performed on the feature association matrix and features are selected to reflect the main relationship and mutual influence between power generation equipment and load clusters in the power system, improve the matching degree and fusion effect, and then build an efficient power generation load fusion model.

[0058] S5. Optimizing the power mutation rate and load change rate corresponding to the historical power generation data and the historical power consumption data respectively by a parameter identification method, and integrating the power generation and load fusion model into a preset generalized load model structure based on the optimization result, to obtain a fusion equivalent model to predict the load demand of the distributed new energy power generation system; In one embodiment, the power mutation rate and load change rate corresponding to the historical power generation data and the historical power consumption data are optimized by a parameter identification method, and the power generation load fusion model is integrated into a preset generalized load model structure based on the optimization result to obtain a fusion equivalent model, including: Extracting historical power mutation rate and historical load change rate from the historical power generation data and the historical power consumption data respectively, and performing sensitivity calculation on the historical power mutation rate and the historical load change rate by using a cross-validation method to obtain a parameter sensitivity index; Taking the parameter sensitivity index as a weight coefficient, optimizing the historical power mutation rate by a recursive least square method with the weight coefficient, and optimizing the historical load change rate according to the parameter sensitivity index; Support vector regression fitting is performed on the optimized historical power mutation rate and the optimized historical load change rate, and a quadratic generalized load expression including a power response term and a load response term is constructed according to the fitting results to obtain the fusion equivalent model.

[0059] Specifically, the present invention uses a quadratic polynomial generalized load function to perform structural mapping on the fusion equivalent model. The generalized load function in the form of a quadratic polynomial contains power terms and cross terms. The power terms reflect the basic load characteristics, and the cross terms reflect the coupling effect. In view of the rapid response characteristics of the distributed new energy power generation system, the power segmentation point is set when the power change rate in the power output curve and the power load curve corresponding to the historical power generation data and the historical power consumption data is greater than 0.3, and the power segmentation point is set, and the historical power mutation rate and the historical load change rate are obtained. The sensitivity of these two mutation parameters is calculated by cross-validation method, and the mean square error optimization function is constructed by a genetic algorithm based on real number coding. The parameter convergence threshold is set to 0.001, and the parameter sensitivity index is obtained by iterative calculation.

[0060] The least squares estimation with weighted coefficients is used to identify the generalized load parameters, and the weight coefficients are determined by the parameter sensitivity index. The historical power mutation rate is optimized by the recursive least squares method. When the parameter change of three consecutive iterations is less than the preset convergence threshold, the iteration is stopped to obtain the optimized historical power mutation rate, and the historical load change rate is updated by the parameter sensitivity index to obtain the optimized historical load change rate. The optimized load change rate and the optimized power change rate are fitted by support vector regression with Gaussian kernel function to construct a quadratic generalized load expression containing power response terms and load response terms, and obtain a fusion equivalent function that characterizes the characteristics of distributed loads to predict the load demand of distributed new energy power generation systems. Among them, in the fitted quadratic generalized load expression, the power response term coefficient is 0.75, the load response term coefficient is 0.45, and the cross term coefficient is 0.35. The fused equivalence function reflects the typical characteristics of the load cluster of the distributed renewable energy power generation system: the power mutation rate dominates the rapid startup phase, the load change rate is more significant during steady-state operation, and the coupling effect between the two is most obvious in the transition phase. It accurately describes the dynamic response characteristics of the distributed load group and provides a basis for the operation optimization of the distributed renewable energy power generation system.

[0061] The present invention can significantly improve the precision and accuracy of the model by accurately evaluating and optimizing parameter sensitivity and constructing a fusion equivalent model using support vector regression fitting; the introduction of the recursive least squares method enables the model to adapt to data changes and the needs of real-time systems, thereby enhancing the adaptability and flexibility of the model; the formed fusion equivalent model can provide strong support for the optimal scheduling and stable operation of the power system, and provide decision makers with more accurate and reliable information and basis. The scheme solves the matching problem between the randomness and intermittency of distributed renewable energy power generation systems and the diversity and time-varying nature of load clusters, realizes the coordinated optimization of renewable energy and loads, accurately reflects the characteristics of distributed renewable energy power generation systems, provides strong support for system planning and operation, and helps to improve the renewable energy absorption capacity and system operation efficiency.

[0062] In the embodiment of the present application, based on the problem of how to improve the prediction accuracy of the load demand of the distributed renewable energy power generation system, a load demand prediction method for the distributed renewable energy power generation system is designed, which obtains the historical data corresponding to the power generation equipment and the load cluster respectively to extract characteristic parameters therefrom, and performs cluster analysis on these characteristic parameters to divide the power generation equipment and the load cluster into different subsets; the historical data corresponding to each power generation equipment cluster subset and the load cluster cluster subset are processed to obtain a clear power generation output curve and a power load curve; and by analyzing the correlation between the power generation data and the power consumption data, the correlation index reflecting the system power generation output and load fluctuation is obtained to fuse the power generation output curve and the power load curve into a power generation load fusion model; the parameter identification method is used to optimize the power mutation rate and load change rate corresponding to the historical power generation and power consumption data respectively, and the power generation load fusion model is integrated into the generalized load model structure based on the optimization result, so as to obtain a fusion equivalent model that can comprehensively consider the changes in renewable energy power generation and load demand, so as to predict the load demand of the distributed renewable energy power generation system, thereby improving the reliability and accuracy of the load demand prediction of the distributed renewable energy power generation system, and providing a strong guarantee for the economic, safe and stable operation of the system.

[0063] It should be noted that although the steps in the above flowchart are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders.

[0064] In another embodiment, if Figure 2 As shown, the second aspect of the present invention provides a load demand forecasting system for a distributed renewable energy power generation system, comprising: A parameter set building module 10 is used to obtain historical power generation data of power generation equipment in a distributed renewable energy power generation system and historical power consumption data of a load cluster, and to extract characteristic parameters from the historical power generation data and the historical power consumption data to build a characteristic parameter set of the distributed renewable energy power generation system; A cluster subset division module 20, configured to perform cluster analysis on the characteristic parameter set, and divide the power generation equipment and the load cluster into a plurality of power generation equipment cluster subsets and a plurality of load cluster cluster subsets according to the cluster analysis result; The subset curve construction module 30 is used to process the specific historical power generation data corresponding to each of the power generation equipment cluster subsets to obtain the power generation output curve of each of the power generation equipment cluster subsets, and to process the specific historical power consumption data corresponding to each of the load cluster cluster subsets to obtain the power consumption load curve of each of the load cluster cluster subsets; A fusion model generation module 40 is used to analyze the correlation between the historical power generation data and the historical power consumption data to obtain a correlation index reflecting the relationship between the power generation output and load fluctuation of the distributed new energy power generation system, and to fuse each of the power generation output curves and each of the power consumption load curves according to the correlation index to obtain a power generation load fusion model; The load demand prediction module 50 is used to optimize the power mutation rate and load change rate corresponding to the historical power generation data and the historical power consumption data respectively through a parameter identification method, and integrate the power generation load fusion model into a preset generalized load model structure based on the optimization result to obtain a fusion equivalent model to predict the load demand of the distributed new energy power generation system.

[0065] It should be noted that each module in the above-mentioned load demand forecasting system for distributed renewable energy power generation systems can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules. For the specific definition of a load demand forecasting system for distributed renewable energy power generation systems, please refer to the above definition of a load demand forecasting method for distributed renewable energy power generation systems. The two have the same functions and effects, which will not be repeated here.

[0066] In summary, the present invention relates to the field of data processing technology, and discloses a load demand prediction method and system for a distributed renewable energy power generation system, which obtains historical data corresponding to power generation equipment and load clusters respectively to extract characteristic parameters therefrom and divides the power generation equipment and load clusters into different subsets by performing cluster analysis; processes the historical data corresponding to each power generation equipment cluster subset and load cluster cluster subset to obtain a power generation output curve and a power load curve; and obtains a correlation index by analyzing the correlation between historical power generation and power consumption data to fuse the power generation output curve and the power load curve into a power generation load fusion model; uses a parameter identification method to optimize the power mutation rate and load change rate corresponding to the historical power generation and power consumption data respectively, and integrates the power generation load fusion model into the generalized load model structure based on the optimization result to obtain a fusion equivalent model to accurately predict the load demand of the distributed renewable energy power generation system.

[0067] Each embodiment in this specification is described in a progressive manner, and the same or similar parts of each embodiment can be directly referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. It should be noted that the technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above-mentioned embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0068] The above-mentioned embodiments only express several preferred implementation modes of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in the technical field, several improvements and substitutions can be made without departing from the technical principles of the present invention, and these improvements and substitutions should also be regarded as the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be based on the protection scope of the claims.

Claims

1. A method for predicting load demand for a distributed renewable energy power generation system, characterized in that: include: Acquire historical power generation data of power generation equipment in a distributed renewable energy power generation system and historical power consumption data of a load cluster, and extract characteristic parameters from the historical power generation data and the historical power consumption data respectively to construct a characteristic parameter set of the distributed renewable energy power generation system; Performing cluster analysis on the characteristic parameter set, and dividing the power generation equipment and the load cluster into a plurality of power generation equipment cluster subsets and a plurality of load cluster cluster subsets respectively according to the cluster analysis result; Processing the specific historical power generation data corresponding to each of the power generation equipment cluster subsets to obtain the power generation output curve of each of the power generation equipment cluster subsets, and processing the specific historical power consumption data corresponding to each of the load cluster cluster subsets to obtain the power consumption load curve of each of the load cluster cluster subsets; Analyze the correlation between the historical power generation data and the historical power consumption data to obtain a correlation index reflecting the relationship between the power generation output and load fluctuation of the distributed new energy power generation system, and fuse each of the power generation output curves and each of the power consumption load curves according to the correlation index to obtain a power generation load fusion model; The power mutation rate and load change rate corresponding to the historical power generation data and the historical power consumption data are optimized through the parameter identification method, and based on the optimization result, the power generation load fusion model is integrated into the preset generalized load model structure to obtain the fusion equivalent model to predict the load demand of the distributed new energy power generation system.

2. A method for predicting load demand for a distributed renewable energy power generation system according to claim 1, characterized in that: The extracting characteristic parameters from the historical power generation data and the historical power consumption data to construct a characteristic parameter set of the distributed new energy power generation system includes: Normalizing and inspecting the historical power generation data and the historical power consumption data respectively to obtain processed historical power generation data and processed historical power consumption data; The power fluctuation, power change rate, intermittent characteristic value and randomness index of the processed historical power generation data are calculated respectively as power generation characteristic parameters; Extracting the periodic variation characteristic value of the processed historical power consumption data based on discrete Fourier transform, and classifying the processed historical power consumption data according to the load type, so as to extract the peak and valley position of the historical power consumption curve corresponding to each type of load, and taking the periodic variation characteristic value as the load characteristic parameter; The radial basis function neural network is trained with the power generation characteristic parameters as input and the load characteristic parameters as output to obtain a characteristic parameter set; the characteristic parameter set is configured as an associated expression of the power generation characteristics and the power consumption characteristics in the distributed new energy power generation system.

3. A method for predicting load demand for a distributed renewable energy power generation system according to claim 1, characterized in that: The cluster analysis is performed on the characteristic parameter set, and the power generation equipment and the load cluster are divided into a plurality of power generation equipment cluster subsets and a plurality of load cluster cluster subsets according to the cluster analysis result, including: The characteristic parameter set is smoothed by a Butterworth low-pass filter, and the hourly mean value, daily power standard deviation and hourly power change rate of the power data in the processed characteristic parameter set are extracted by a sliding window to construct a power generation index sequence; Based on the probability distribution characteristics of the power generation index sequence according to quantile statistics, the Euclidean distance between each of the power generation equipment is calculated through hierarchical clustering according to the probability distribution characteristics to compare with the distance threshold, and each of the power generation equipment is divided according to the comparison result to obtain a plurality of power generation equipment cluster subsets.

4. A method for predicting load demand for a distributed renewable energy power generation system according to claim 3, characterized in that: The cluster analysis is performed on the characteristic parameter set, and the power generation equipment and the load cluster are divided into a plurality of power generation equipment cluster subsets and a plurality of load cluster cluster subsets according to the cluster analysis result, and further includes: Performing wavelet decomposition on the characteristic parameter set to obtain the load peak duration, peak-to-valley difference value and load root mean square error value of each load cluster at different time scales to construct a load characteristic vector; A self-organizing map network is used to cluster the load feature vector, the number of nodes in the competition layer of the self-organizing map network is used as the preset clustering number, and the similarity between the input vector and the weight vector of the self-organizing map network is calculated by the Euclidean distance between each load cluster, so as to update the weights of the winning node and its neighboring nodes in the self-organizing map network, and obtain several load cluster clustering subsets.

5. A method for predicting load demand for a distributed renewable energy power generation system according to claim 1, characterized in that: The specific historical power generation data corresponding to each of the power generation equipment cluster subsets is processed to obtain the power generation output curve of each of the power generation equipment cluster subsets, including: The historical power generation data of each power generation device contained in each of the power generation device cluster subsets is used as the specific historical power generation data of the corresponding power generation device cluster subset, and each of the specific historical power generation data is normalized to obtain a standard power time series data set of each of the power generation device cluster subsets; Performing a three-layer decomposition on each of the standard power time series data sets by using a wavelet basis function to extract the power fluctuation characteristics of each of the power generation equipment cluster subsets, and extracting the principal components in the power fluctuation characteristics based on a singular spectrum analysis method to obtain the power fluctuation characteristic value of each power generation equipment in each of the power generation equipment cluster subsets; According to each of the power fluctuation characteristic values ​​and their corresponding normal power generation condition Gaussian distribution components and intermittent power generation condition Gaussian distribution components, a mixed Gaussian function is constructed, and the distribution parameters of the mixed Gaussian function are solved based on the maximization of the log-likelihood function, so as to describe the transition law of the power generation equipment between multiple power generation states through a Markov chain, and obtain the random power generation characteristic parameters of each of the cluster subsets of the power generation equipment; Quantifying the power mean of the power generation equipment contained in each of the power generation equipment cluster subsets, and quantifying the root mean square error between the specific historical power generation data of a single power generation equipment in the same power generation equipment cluster subset and the power mean, so as to use the power generation equipment with the smallest root mean square error as the representative power generation equipment of the corresponding power generation equipment cluster subset; The radial basis kernel function is used to interpolate and fit the specific historical power generation data and corresponding random power generation characteristic parameters of the representative power generation equipment in each of the power generation equipment cluster subsets to obtain the power generation output curve of each of the power generation equipment cluster subsets.

6. A method for predicting load demand for a distributed renewable energy power generation system according to claim 1, characterized in that: The specific historical power consumption data corresponding to each of the load cluster subsets is processed to obtain the power consumption load curve of each of the load cluster subsets, including: The historical power consumption data of each load cluster contained in each of the load cluster cluster subsets is used as the specific historical power consumption data of the corresponding load cluster cluster subset, and each of the specific historical power consumption data is season-marked, workday attribute-marked, and normalized to obtain a standard load time series data set of each of the load cluster cluster subsets; Performing three-layer wavelet decomposition on each of the standard load time series data sets to obtain load fluctuation characteristics on multiple time scales, and performing peak detection on each of the standard load time series data sets to obtain the start and end positions and duration of the load peak period, so as to construct the load shape feature vector of each of the load clustering subsets with the load fluctuation characteristics; The support vector regression of the radial basis kernel function is used to calculate the similarity between the characteristic vectors of each load form as a similarity index, and the correlation index between each load cluster is quantified based on the overlap of the start and end positions and duration of each load peak period, so as to construct a diversity feature set of each standard load time series data set with the similarity index; Based on each of the diversity feature sets, a recursive neural network with a double-layer cyclic structure is used to establish the power load curve of each of the load cluster cluster subsets; the first cyclic hidden layer of the recursive neural network is used to extract the load morphological characteristics of each of the load cluster cluster subsets, and the second cyclic hidden layer of the recursive neural network is used to extract the time-varying law of power consumption of each of the load cluster cluster subsets.

7. A method for predicting load demand for a distributed renewable energy power generation system according to claim 1, characterized in that: The analysis of the correlation between the historical power generation data and the historical power consumption data to obtain a correlation index reflecting the relationship between the power generation output and load fluctuation of the distributed new energy power generation system includes: The historical power generation data and the historical power consumption data are processed respectively to obtain a historical power generation sequence and a historical load characteristic sequence, and the historical power generation sequence and the historical load characteristic sequence are time-aligned using a time sliding window to calculate the Pearson correlation coefficient in each sliding window, and a fluctuation synchronization curve for characterizing the correlation between power generation and load changes is constructed based on the correlation coefficient calculation result; Quantify the fluctuation trend characteristic value of the fluctuation synchronization curve within a preset time interval, and perform multi-scale analysis on the historical power generation sequence and the historical load characteristic sequence through three-layer wavelet decomposition to obtain a multi-scale sequence entropy value; Based on the multi-scale sequence entropy value and the fluctuation trend characteristic value, a correlation characteristic vector is constructed, and the correlation characteristic vector is used as a correlation index reflecting the power generation output and load fluctuation of the distributed new energy power generation system.

8. A method for predicting load demand for a distributed renewable energy power generation system according to claim 1, characterized in that: The step of fusing the power generation output curves and the power load curves according to the associated index to obtain a power generation load fusion model includes: Performing joint dimension reduction and restoration reconstruction processing on each of the power output curves and each of the power load curves through a codec to obtain a power generation load fusion feature to construct a feature association matrix with the association index; Performing singular value decomposition on the feature association matrix to obtain a feature subspace, and selecting basis vectors corresponding to feature association values ​​whose cumulative contribution rates reach a preset contribution threshold from the feature subspace as matching features, so as to calculate the matching degree between each of the power generation equipment cluster subsets and each of the load cluster cluster subsets; The power generation output curves of the power generation equipment cluster subset whose matching degree exceeds a preset matching threshold are fused with the power consumption load curves of the load cluster cluster subset to obtain the power generation load fusion model.

9. A method for predicting load demand for a distributed renewable energy power generation system according to claim 1, characterized in that: The power mutation rate and load change rate corresponding to the historical power generation data and the historical power consumption data are optimized by the parameter identification method, and the power generation load fusion model is integrated into the preset generalized load model structure based on the optimization result to obtain the fusion equivalent model, including: Extracting historical power mutation rate and historical load change rate from the historical power generation data and the historical power consumption data respectively, and performing sensitivity calculation on the historical power mutation rate and the historical load change rate by using a cross-validation method to obtain a parameter sensitivity index; Taking the parameter sensitivity index as a weight coefficient, optimizing the historical power mutation rate by a recursive least square method with the weight coefficient, and optimizing the historical load change rate according to the parameter sensitivity index; Support vector regression fitting is performed on the optimized historical power mutation rate and the optimized historical load change rate, and a quadratic generalized load expression including a power response term and a load response term is constructed according to the fitting results to obtain the fusion equivalent model.

10. A load demand forecasting system for a distributed renewable energy power generation system, characterized in that: include: A parameter set building module, used to obtain historical power generation data of power generation equipment in a distributed renewable energy power generation system and historical power consumption data of a load cluster, and extract characteristic parameters from the historical power generation data and the historical power consumption data to build a characteristic parameter set of the distributed renewable energy power generation system; A cluster subset division module, used for performing cluster analysis on the characteristic parameter set, and dividing the power generation equipment and the load cluster into a plurality of power generation equipment cluster subsets and a plurality of load cluster cluster subsets respectively according to the cluster analysis result; A subset curve construction module is used to process the specific historical power generation data corresponding to each of the power generation equipment cluster subsets to obtain the power generation output curve of each of the power generation equipment cluster subsets, and to process the specific historical power consumption data corresponding to each of the load cluster cluster subsets to obtain the power consumption load curve of each of the load cluster cluster subsets; A fusion model generation module is used to analyze the correlation between the historical power generation data and the historical power consumption data to obtain a correlation index reflecting the relationship between the power generation output and load fluctuation of the distributed new energy power generation system, and fuse each of the power generation output curves and each of the power consumption load curves according to the correlation index to obtain a power generation load fusion model; The load demand prediction module is used to optimize the power mutation rate and load change rate corresponding to the historical power generation data and the historical power consumption data respectively through the parameter identification method, and integrate the power generation load fusion model into the preset generalized load model structure based on the optimization result to obtain the fusion equivalent model to predict the load demand of the distributed new energy power generation system.

Images