Study Of Short-term Wind Speed Prediction Method

From1980s, wind power generation has developed more and more rapidlyaround the world. Significantly different from traditional power generated by hydro orthermal plant, output of wind generator is intermittent, random, fluctuating anduncertain. It will bring a lot of stable and qualitative challenges to power system. As akey technique, short-term prediction of wind farm is deemed to be an effective way tosolve the problems.Wind power prediction can help operators to make dispatch plans which can beused to coordinate various power sources in power system, reduce the negative impactof wind output to the power grid and improve the economic performance of powersystem operations. In this paper, some aspects of short-term prediction for wind speedand wind power are studied and discussed:1) A wind speed prediction method based on BFNN (basis function neuralnetwork) is proposed. In the given method, a suitable model for wind speed predictionis built using the Chebyshev neural network. Time series of historical wind speeds arethen used to train the network so as to predict the wind speed in the future. Simulationresults based on real wind farm data show that the given method is better than thetraditional one, named persistence method. Compared with traditional BP neuralnetwork, BFNN has lower dimension and faster calculation speed so that it canimprove the prediction efficiency.2) A method to improve the predictive error is proposed. The method has twosteps. In step1, it estimates the future predictive error by using the previous predictiveerror list. Then, non-parametric kernel density estimation is used to reduce the futurepredictive error. In step2, numerical weather prediction (NWP) is embedded tofurther improve the precision of the prediction result. It is found that the given methodcan improve the predictive precise by using real data given by some real wind farms.3) A method to correct wind speed prediction based on NWP is proposed finally.It determines similar days of the predictive one, and uses their predictive errors torevise predicting result. Influence of the number of similar days and the value of thecorrection coefficient to the method is discussed in the study. Simulation resultsreveal the effectiveness and correctness of the given method.