加权多分位鲁棒ELM的短期负荷预测方法  

Weighted Multi-quantile Robust ELM for Short-term Load Forecasting

作　　者：鲁迪 王星华[1] 刘升伟 陈豪君 贺小平[1] LU Di;WANG Xinghua;LIU Shengwei;CHEN Haojun;HE Xiaoping(School of Automation,Guangdong University of Technology,Guangzhou 510006,China)

机构地区：[1]广东工业大学自动化学院,广州510006

出　　处：《电力系统及其自动化学报》2020年第3期33-38,共6页Proceedings of the CSU-EPSA

基　　金：国家自然科学基金资助项目(51707041);中国南方电网公司科技项目(GDKJXM20162087)。

摘　　要：为获取足够精确的短期负荷预测值作为电力系统规划和运行的依据,提出一种加权多分位鲁棒极限学习机ELM(extreme learning machine)的短期负荷预测方法。首先融合分位回归与鲁棒ELM形成多分位鲁棒ELM基本预测模型,然后通过选取不同的分位值来模拟所有的可能性预测场景,以此得到不同分位场景下的预测值。最后按照“误差大、权值小;误差小、权值大”的误差反馈加权原则对上述不同分位下的预测值进行加权求和,以此得到最终的预测结果。实例证明该混合模型预测方法适用性强,且能取得较高的预测精度。To obtain short-term load forecasting results with an accuracy high enough for the planning and operation of a power system,a weighted multi-quantile robust extreme learning machine(WMQ-RELM)is proposed for short-term load forecasting.First,quantile regression and robust ELM are combined to form a basic forecasting model based on multi-quantile RELM.Then,different values of quantile are selected to simulate all the possible prediction scenarios,thus obtaining the prediction results under scenarios with different quantiles.At last,the prediction results with differ ent quantiles are summed with weights according to the error feedback weighting principle,i.e.,a large prediction error for a small weight and a small one for a large weight.The results of a case study show that the proposed forecasting meth od based on a hybrid model is of strong applicability and can achieve a higher prediction accuracy.

关 键 词：短期负荷预测 加权多分位鲁棒极限学习机 误差反馈加权 

分 类 号：TM715[电气工程—电力系统及自动化]