分位数回归在风电功率预测不确定性分析中的应用

QUANTILE REGRESSION IN UNCERTAINTY ANALYSIS OF WIND POWER FORECASTING

基于分位数回归原理定义风电功率预测风险指数——PaR(Predict at Risk),并针对不同预测模型的不确定性因素来源分别建立短期和超短期预测的不确定性分析模型。该模型可提供在任意置信水平下,预测功率可能出现的波动范围。将该模型应用于中国北方某风电场进行风电功率短期及超短期预测的不确定性分析,实验结果表明:较已有不确定性分析方法,该方法无需假设预测功率误差分布,既适用于基于历史数据的预测方法也适用于基于数值天气预报的预测方法,且计算过程简单。

A risk index of wind power prediction was defined which was termed as PaR （Prediction at Risk）. As different forecasting model triggering different uncertain factors, uncertainty analysis models were built for two prediction models which were short term prediction model based on NWP and ultra short term prediction model based on historical data. These uncertainty analysis models would provide potential fluctuation range of predictive power under any confidence level. The proposed models were applied to a wind farm in north China and the results show that the model needs a little calculation cost and needn＇ t supposing forecasting error distribution comparing with existing uncertainty analysis methods. Besides, this method can not only be utilized for historical data based forecasting model, but also NWP-based forecasting model.