Study On The Short-term Wind Power Forecasting In Wind Farm

This paper mainly research on the shortcomings of the accuracy ofprediction, reliability and adaptability of the short-term wind powerprediction technique and it focuses on reaching to improve the predictionaccuracy and reliability of the algorithm. The main contents are asfollows:As many domestic researches have not adequately considered theimpact of numerical weather factors, not fully exploit the information ofthe data it contains. For this phenomenon, this paper proposes localoutlier detection method based on density clustering, and improves themethod of selecting similar day data bases on the traditional method.Then combine time series to create a single forecasting model. Thesimulation results show that the method of this paper proposed is feasibleand effective.This paper proposes new combinations according to the combinationof traditional time-series model and BP neural network model. Thenmaking the new model compare with other models, the results show theprecision of the forecast are improved.For the result of the past short-term wind power prediction method isa single point, this paper introduces the concept of horizontal and verticalerror, proposes a new interval estimation model bases on horizontal andvertical error, proposes the selected method of horizontal and verticalerror weighting, establishes a new interval estimation model bases onhorizontal and vertical error. Simulation experiments show that the newmodel is better than the traditional model in practicality and has bettervalue on risk assessment.The prediction method studied in this paper is applied to thedevelopment of predictive systems. The system can not only help griddepartments to develop grid scheduling but also provide a reference forthe wind farm personnel arrangements for working hours.