Research On Wind Speed Prediction Based On Correlation Analysis And Machine Learning

With the constant change of global energy form and demand,China’s energy production structure has undergone great changes.The key to successfully achieving the ambitious 30·60 carbon neutral goal is to actively develop renewable energy to replace polluting fossil fuels.Wind energy with its clean,pollution-free characteristics,has become one of the most potential renewable energy.However,the intermittent and fluctuating characteristics of wind energy itself cannot be ignored.In the process of grid connection,it may have a great impact on the stability of power grid and become an important factor restricting the development of wind industry.To improve the accuracy of wind speed prediction is one of the keys to solve the above problems.Most of the existing wind speed prediction methods focus on model improvement,which lacks consideration of the spatio-temporal correlation of wind speed of wind turbines and ignores the autocorrelation of wind speed and the interaction between wind turbines,thus affecting the overall effect of wind speed prediction.How to correctly analyze the correlation of wind turbine and improve the accuracy of forecast is an urgent problem for wind power industry.Based on this,this paper firstly introduces the formation process and variation rule of wind speed in detail,and clarifies the cause of the complex space-time correlation of wind speed.By comparing several commonly used wind speed forecasting methods,it is found that the current forecasting methods lack time correlation.Therefore,the Long short-term Memory(LSTM)network model in machine learning is used to predict wind speed series.The simulation results show that the LSTM network model can effectively extract the time dependence of wind speed data and improve the prediction accuracy to a certain extent.Then,in view of the spatial correlation problems existing in LSTM model,the paper proposes a method of using Copula function to analyze the spatial correlation of wind speed,and combining LSTM network model with Copula function to predict the wind speed.In order to ensure the accuracy of Copula function analysis results,based on Copula function theory,this method uses Kernel Density Estimation(KDE)to estimate the wind speed edge distribution function and Copula function parameters.The optimal Copula function which can accurately describe the interaction between fans is determined by goodness of fit test.By combining the Copula function correlation analysis with LSTM prediction,a single Copula function prediction model was constructed and simulated,which verified the superiority of Copula function in correlation analysis,and further indicated that the prediction accuracy can be effectively improved by analyzing the spatio-temporal correlation and establishing a prediction model based on spatio-temporal correlation analysis.Finally,a hybrid Copula model combining multiple Copula functions is proposed to accurately fit multiple wind turbines in a wind farm.In order to ensure the fitting effect of the mixed Copula function,the maximum expectation(EM)algorithm is introduced to calculate the weight coefficients and dependence coefficients of the mixed Copula function model.By comparing the goodness of fit test results of the mixed Copula function with that of the single Copula function,the improved correlation analysis of the hybrid Copula function is verified.On this basis,combined with the time correlation distribution process of LSTM prediction model,the hybrid Copula function prediction model is constructed.The experimental results show that the hybrid Copula function can combine the advantages of multiple Copula functions and further analyze the correlation of wind speed.It further proves that accurate correlation analysis is the key to improve the accuracy of wind speed prediction.