Research On Forecasting Model Of Wind Speed In Wind Farm

On wind speed mutations, the effect of traditional wind speed forecasting method is poor. Therefore, in-depth studies of the changes of wind speed sequence is necessary, appropriate methods are selected to reduce the prediction error. Conventional prediction methods belong to deterministic prediction methods. Considering randomness of wind speed, Studying wind speed forecasting error distribution from the point of probability statistics is more meaningful.The main work and innovations are as follows:(1) Wave characteristics of wind speed are quantitative analyzed; Comparing BP neural network, the GRNN and conventional SVM, results show that the conventional method has poor ability of tracking wind speed mutation.(2) According to different characteristic scales, the wind speed time series are decomposed by LMD to reduce the nonstationarity and excavate internal characteristic information of wind speed. Traditional LMD algorithm employs the moving average method to obtain the local mean function and local envelope function, which causes over-smoothness easily and affects precision of decomposition. Therefore, Akima Interpolation Method is proposed to improve LMD. And a set of evaluation index is put forward.(3) Due to LS-SVM dependents on nuclear function and parameter selection, Segmented Kernel Least Squares Support Vector Machine(SK-LSSVM) is established. Considering least squares criterion expands fitting error of wind anomaly points, PSO based on least-absolute criteria is adopted to parameters optimization. Then the combination forecast model is built up based on the improved LMD and SK-LSSVM.(4) Study the cause of the wind speed prediction error from the point of probabilistic, analysis the distribution characteristics of wind speed prediction error by statistical. Mixed skew distribution and other distributions are proposed to apply to estimate distribution model of short-term wind speed prediction error.