考虑多位置NWP和非典型特征的短期风电功率预测研究

Research of Short-Term Wind Power Forecasting Considering Multi-Location NWP and Uncanonical Feature

数值天气预报（numerical weather prediction,NWP）是短期风电功率预测模型的主要输入。通常,传统模型只考虑NWP的风速、风向、温度、湿度、压强这5类典型特征,且多数在单位置NWP的基础上建立。因此,为充分利用NWP信息,研究了NWP非典型特征的可用性,并考虑了多个位置的NWP信息。在考虑多位置NWP及非典型特征时,提出了以最大相关-最小冗余原则提取输入变量的预测方法,并和通过主成分分析提取的方法进行对比。结果表明,多位置NWP和非典型特征均包含有效信息,有利于提高预测精度。而在考虑多位置NWP和非典型特征时,以最大相关-最小冗余原则建立的模型比通过主成分分析建立的模型预测精度更高,其均方根误差较只考虑单位置NWP和典型特征时降低了1.84%。

Numerical weather prediction（NWP） is the main input of short-term wind power forecasting model. Usually, only five canonical features in the NWP, namely wind speed, wind direction, temperature, humidity and pressure, are considered in conventional models, most of which are established based on the NWP of single location. In order to make full use of NWP information, applicability of uncanonical features is investigated and the NWP information of multiple locations is considered. By combining the NWP information of multiple locations with uncanonical features, the method where models are established based on input variables obtained with the principle of minimal redundancy and maximal relevance is proposed and compared with the method based on principal component analysis. Results show that NWP of multiple locations and its uncanonical features contain useful information, beneficial to improve prediction accuracy. By considering the NWP of multiple locations and uncanonical features, the forecasting model based on principle of minimal redundancy and maximal relevance shows advantages over the model based on principal component analysis and improves its RMSE by 1.84%, compared with the model only considering NWP of single location with canonical features.