Short-Term Forecast Based on Generalized Maximum Correntropy Criterion and Kernel Extreme Learning Machine

Abstract

How to realize the short-term forecast efficiently and accurately is of great significance to the stable economic operation of electric power system and the operation management of electric power company. Traditional short-term forecast method usually takes a long time to process data and is easy to occur errors. In this paper, a short-term forecast method based on generalized maximum correntropy criterion and kernel extreme learning machine is proposed. More precisely, multidimensional features representing the data characteristics are selected, date features and k-means clustering are carried out to obtain a power data cluster center. Then, generalized maximum correlation entropy is used to filtrate different users with similar power characteristics, and construct the advanced training data set. Finally, a novel kernel extreme learning machine model is used to simulate and output the short-term load prediction. Experimental results and comparison on real data show that our method can effectively improve the accuracy of prediction and has strong robustness.