Multi-wind Farm Power Prediction Based On Spatio-temporal Feature Extraction And Data Dimensionality Reduction

The depletion of fossil energy and the deterioration of the ecological environment have promoted the development of renewable energy to become a global consensus.As far as wind power is concerned,due to the time-varying volatility,intermittency and uncertainty of wind energy,as well as the complex spatial correlation between multiple wind farms in the region,it is currently difficult to accurately grasp its power generation rules,thereby threatening the safety and stability of the power system run.How to effectively grasp the laws of the temporal and spatial correlation of wind power and improve the power grid’s ability to accept wind power is still facing a dilemma.Based on the distribution characteristics of domestic wind power bases,this paper uses spatio-temporal feature mining as clues,and spatio-temporal feature extraction methods as a means to carry out research on wind power forecasting for regional multi-wind farms,which has important theoretical significance and engineering value for grasping the causes of wind power changes and coping strategies.Based on the complex and coupled spatio-temporal correlations of multiple wind farms in the region,this paper uses a variety of feature extraction techniques to propose a short-term power prediction method for multi-wind farms with spatio-temporal feature extraction and data dimensionality reduction.The main work and contributions of the thesis are as follows:Firstly,based on the description of the research background,significance and research necessity,the Pearson correlation coefficient and the maximum mutual information coefficient are used to analyze the spatial correlation of multi-wind farms,and the autocorrelation coefficient and cross-correlation coefficient are used to analyze the time series correlation and spatio-temporal coupling correlation of multiple wind farms.Based on the conclusion of correlation analysis and the "dimension disaster" problem encountered in wind power prediction of multi-wind farms,a spatial multi-dimensional wind power reduction and reconstruction method based on the optimal RBF core principal component is proposed.Secondly,based on the advantages of deep learning for feature extraction of massive data,a deep prediction framework of dimensionality reduction coding-feature prediction-reconstruction decoding is proposed to achieve power feature extraction of multiple wind farms,dimensionality reduction and feature optimization of potential influencing factors,the independent prediction of power characteristics and the organic unity of reconstruction and restoration.Finally,in view of the characteristics of the spatio-temporal correlation of coupling between multiple wind farms,a graph neural network model architecture based on the adjacency matrix is proposed to make short-term and ultra-short-term forecasts of wind power from multiple wind farms.Based on the actual data of a domestic wind power base,the analysis and verification of a calculation example show the validity and correctness of the ideas and methods in this paper.