Wind energy is one of the most rapidly growing energy resources all over the world.Due to high penetration rates of wind energy generation,the problem about the stability of power system becomes important.Wind power prediction for a few hours ahead is also an essential process for power system generators planning and scheduling.So it increases the necessity for more accurate and reliable techniques for wind power generation prediction.Accurate forecasting of wind speed is usually the first step to wind power prediction.Application of wavelet de-noising to the analysis of wind speed sequence,the paper presents an advanced method for wind speed forecasting three hours ahead.It is a combination forecasting model on Least squares support vector machines(LS-SVM) and time series. Firstly,the wind speed sequence can decompose to trend component(low-frequency part) and stochastic component(high-frequency part) by Wavelet Transform.Secondly the trend component can be searched for similar sequence from historical data.Based on LS-SVM for regression,the trend component can be forecasted three hours ahead.Stochastic component belong to stationary time series,so it is suitable for multi-step forecasting by Time Series Analysis.The results of wind speed forecasting for Changling wind farm reveal the effectiveness and the accuracy of the proposed models.With an improvement over the persistent and BP neural network,the proposed models are more suitable for multi-step wind speed forecasting.
|