Research On Effective Processing Of Data And Samples Of Wind Speed Forecasting And Its Model Optimization

With the aggravation of energy crisis and environment pollution, wind energy has been paid widespread attention by reason that it’s clean, widely distributed and renewable. Wind power generation as the main means of using wind energy has been developing rapidly and the proportion of wind power in the grid is also increasing. However, the randomness and the intermittent of wind lead to the instability of wind power, and then the uncontrollability of wind power is increased. And the serious threat will be brought to power system when bringing wind power into the grid. Therefore, the wind speed must be accurately forecasted to improve the controllability of wind power and ensure the stable operation of power system.In view of the existing problems in modeling process of wind speed forecasting, some researches on the preprocessing of wind data, the optimal selection of training samples of neural network and the optimization of wind speed forecasting model are conducted in this paper. The main research of the paper is as follows:(1) Expound the research background and significance of wind speed forecasting, summarize the classification of the wind speed forecasting and the characteristics of common forecasting methods, give the evaluation index of wind speed forecasting results, and summarize the existing problems in modeling process of wind speed forecasting.(2) Taking the actual wind data as the example, firstly, the modified checking standard of wind data is used to find bad data from the wind data, and analysis the different existing situation of bad data according to the checking results. Then the piecewise moving mean substitution, the support vector regression imputation and the combination of piecewise moving mean substitution and support vector regression imputation are applied to deal with bad data under different existing situation. Simulation results show that these suggested methods can obtain better results for bad data under different existing situation.(3) Aiming at the problems of selecting training samples and optimizing hidden layer structure of neural network, the optimal selection method of training samples based on fuzzy clustering and random sampling and the optimization method of hidden layer structure based on grey correlation-contribution pruning method are proposed. Then the wavelet neural network wind speed forecasting model based on selecting training samples and optimizing structure is built, and the GABP neural network wind speed forecasting model based on selecting training samples and optimizing structure is built. Simulation results show that using selected training samples to train neural network can effectively improve the accuracy of wind speed forecasting results; and using grey correlation-contribution pruning method to optimize the hidden layer structure can not only simplify structure of neural network, but also significantly improve the performance of wind speed forecasting model.(4) There are some problems in modeling process of Kalman filtering algorithm, such as the state equation and observation equation is difficult to establish, how to select observed values and so on. So the unitary time series and wavelet neural network are introduced into the modeling of wind speed forecasting based on Kalman filtering algorithm. First, the unitary time series model is built and transformed into state space to get the state equation. Then wavelet neural network model was built by using wind speed sequence, and the results and error of wind speed forecasting are respectively taken as the observed value and observed error, so the observation equation is obtained. Finally, the recursive equations of kalman filter algorithm are used to forecast wind speed. Simulation results show that the forecasting performance of the wind speed forecasting model based on kalman filtering algorithm with proposed method is improved effectively, and the forecasting values of wind speed are more accurate.