Short Term Prediction Of Wind Power Based Nonlinear Time Series And Neural Network

With the increasing proportion of wind power in the world of energy and the rapid development of wind power industry, the high-precision prediction of wind power improves its competitiveness in the electricity market with its effective reduce and avoidness of the bad effects caused by the integration of wind power on the power system, as well as its conduciveness to the implementation of the power grid dispatching and management of wind farm. Historical datas are used in short-term wind power prediction, whereas long time prediction must use numerical weather prediction. Currently, countless measures are used in wind speed and power prediction, among which time series and neural network method in statistical methods are in widespread use. ARIMA model in time series and neural network model are adpoted, combined with the numerical weather prediction, the method of wind speed combination forecasting, and power prediction are proposed. The main work includes the following aspects:According to the relevant research literature, the paper summarizes the development background, realistic meaning and current situation of the prediction of the wind speed and power, as well as its existing problems and difficulties in present research status. The influential meteorological parameters of wind power prediction are briefly introduced, their influence of wind power the processing methods and rules of parameters are analyzed, meanwhile the classification and the principle of the existing prediction methods are presented in detail.ARIMA model, BP neural network model and numerical weather prediction model are built. The law of error analysis of the three methods are analyzed. The wind velocity combination forecast based on entropy is carried out based on the three single models. In order to improve the precision of prediction model, the weight is extended to 24 hours of 96 time sequence of weights based on statistics of prediction error in the three single models. Meanwhile, the sequence weights are updated to obtain a dynamic entropy weight combined forecast model with sequence weights updated by day by adding the obtained forecast error to the desired model.Neural network prediction model of power is established, and the select of the input parameter and the number of neurons in hidden layer under the same conditions are compared and analyzed to determine the best number of input parameters for the optimal power prediction model. Meanwhile, the established wind speed prediction model with dynamic combination is applied to the power prediction mode, The prediction error analysis before and after the application are presented at last.