Short-term Wind Power Prediction Based On Nonparametric Method

The accurate wind power prediction helps reduce the influence of wind power output on power grid. But the study of wind power prediction has not reached a satisfactory level. The methods of wind power prediction in existence mostly give deterministic results only. It is hard to meet the requirements of the uncertainty risk analysis and decision-making in power grid dispatch and electricity trades. Therefore it is necessary to give the probabilistic forecast of wind power.This paper studies the deterministic and probabilistic predictions of short-term wind power based on the nonparametric theory. On one hand, the paper sets up the model of the short-term wind power determinate prediction based on the nonparametric regression method. Firstly, divide the wind power samples corresponding to sampling times, and then collect the wind power of the same time every day of the historical data to constitute a set of wind power time series, such as96points sampling times can be divided into96groups of time series. Secondly, set up the nonparametric autoregressive prediction model to each wind power time series. And then predict the power to the next point of every time series. Thus, the wind power prediction of the next day is obtained. This method just uses the historical data, can avoid the influence of subjective factors, guaranteeing the objectivity of prediction accuracy. Take a case of all wind farms in one province as a whole to predict the total wind power. The prediction outcomes verify the effectiveness of the nonparametric regression model for the wind power short-term prediction.On the other hand, this paper studies the probabilistic forecasts of wind power based on nonparametric estimation. Firstly, this paper discusses and analyzes the model of kernel density estimation. This method has the problem that it requires many sample data but the available historical data in actual doesn’t meet its requirements. To solve the problem, this paper combines the Bootstrap of nonparametric method with kernel estimating method, and establishes the corresponding model of wind power probability prediction. The model resamples the historical data of prediction error based on the Bootstrap method to format a large number of new samples of the prediction error for statistical analysis, and then models the probability density of the prediction error based on the kernel estimating method. Thus it can get the probabilistic prediction results after the results of deterministic prediction of wind power. This article still takes the wind power in one province for example analysis, the results of the numerical example shows that the probabilistic prediction can give the fluctuation range of wind power in the future period compared with the deterministic prediction, which can effectively help the traders to make a decision.