Research On Short-term Forecast Of Wind Power Based On Combination Prediction

The energy crisis has led to an energy revolution.Wind power is developing rapidly and has become an important part of the power energy structure.The randomness and intermittency of wind power threaten the balance and stability of the real-time supply and demand of the power system,thus limiting the power grid’s consumption of wind power.To forecast the short-term wind power is helpful for the load dispatch department to better control the real-time changes of wind power so as to adjust the dispatching plan in time,thus ensuring the stable operation of the power system.The reserve capacity of system can also be arranged more reasonably to reduce the running cost.Therefore,it is of great practical value to forecast the output power of wind farm.Under this background,this paper used time series method,neural network and combined forecasting method to forecast the short-term wind power.Auto Regressive and Moving Average(ARMA)model and the improved clustering ARMA model were established on the basis theory of time series method.An BP neural network model was established on the basic theory of neural network method.The prediction results and characteristics of each model were analyzed by example.Two kinds of combined prediction model were established on the basic theory of combination prediction.One of the models combined clustering ARMA and BP neural network model on the basis theory of linear combination method.The other one used wavelet analysis to combine the clustering ARMA and BP neural network model to form the wavelet-temporal-BP neural network model.The prediction results and characteristics of two combined models are analyzed by example.By comparing the prediction results of various models,it has found that the prediction results of the clustering ARMA model is better than the ARMA model,which indicates that the classification of the sample data can improves the prediction accuracy.The prediction results of the combined model is better than the single model,which shows that the combination of single model can improves the prediction accuracy.The wavelet-temporal-BP neural network model has the highest prediction accuracy,which shows that it’s more suitable for short-term wind power forecasting.