Application Of Realtion Vector Machine Theory In Real-time Prediction Of Wind Power

Wind power as the current best carbon reduction effect of clean energy,is China's current technology under the condition of the most potential for large-scale development of hydropower and renewable energy.But at the same time the wind of its own characteristics,including the randomness and uncertainty,the large-scale wind power grid has some difficulties.In order to realize the large-scale wind energy development and utilization,this article is set in short-term wind power prediction,based on the measured data of multiple wind farms,the jilin province respectively from four aspects,including: wind power data is lacking,the multi-step prediction rolling method,the uncertainty of wind power prediction analysis and prediction error nonparametric fitting has carried on the comprehensive analysis.For predicting the feature analysis of the wind power generation,power and energy storage configuration studies need to be done on the basis of historical data,but in practice often due to various reasons lead to incomplete data,missing data could lead to chaos system is difficult to control,or there are more and more uncertainty,these cases will be on the analysis of the subsequent estimates caused great obstacles.Based on the minimal redundancy maximal relevance principle to swallow the wind wind power data,this paper analysis the variables related to the power,then according to the mutual information theory,the variable by the principle of maximum correlation minimum redundancy feature selection,the connection between the mining characteristics and power,finally according to the link to swallow the power data.Results show that the feature selection is an effective way for dimensionality reduction of high-dimensional data,from the original feature concentration to select feature subset,retain the original feature set of effective information,can be instead of the whole power of wind turbines.Multi-step rolling of wind power prediction model is established.Wind power prediction accuracy is higher,the higher the utilization rate of wind energy,therefore,requires accurate prediction of wind power.Relevance vector machine(RVM)is a kind of sparse probability model of machine learning,have good generalization ability to learn,can effectively predict the wind power and running time very quickly.At the same time introducing collection empirical mode decomposition(EEMD),will the initial power data sequence is decomposed into several group of smooth sequence,this method can significantly improve the prediction accuracy,speed up the running time.Analyze the uncertainty of wind power prediction.Because any forecast has the uncertainty,therefore with confidence interval range of single point prediction can reduce the risk of power grid and wind farm operation,the operation of the whole system are more security and stability.To predict the uncertainty analysis of predictive power of a single value can be converted into success rate estimated interval.Results show that the prediction model can provide relevance vector machine forecast range under the given confidence level.The prediction error fitting distribution evaluation.By analyzing the distribution features of the prediction error can be concluded that the parameter estimation and forecasting method,predict the time interval,the shape of the prediction error probability distribution,and the relationship between the installed capacity of wind farms,which make the system stable running continuously.Results show that the parameter estimation distribution model for the distribution of different scale of wind farm and various conditions can better fitting,including unimodal and bimodal situation,but the unimodal better fitting effect.