Short‐term load forecasting of Australian National Electricity Market by an ensemble model of extreme learning machine. Issue 4 (1st April 2013)

• Record Type: Journal Article
• Authors: Zhang, Rui; Dong, Zhao Yang; Xu, Yan; Meng, Ke; Wong, Kit Po
• Is Part Of: IET generation, transmission & distribution. Volume 7:Issue 4(2013)
• Page Start: 391
• Page End: 397
• Publication Date: 2013-04-01
• Subjects: gradient methods -- learning (artificial intelligence) -- load forecasting -- neural nets -- power engineering computing -- power markets; short‐term load forecasting -- Australian National Electricity Market -- ensemble model -- extreme learning machine -- power system operations -- artificial neural network -- ANN -- STLF problem -- gradient‐based learning algorithm -- ELM learning technique -- stability problem -- forecasting accuracy -- Australian NEM -- pre‐defined range -- training‐tuning speed -- electricity load variation -- NEM historical load data -- training speed; Electric power production -- Periodicals; Electric power transmission -- Periodicals; Electric power distribution -- Periodicals; 621.3105
• DOI: 10.1049/iet-gtd.2012.0541
• Languages: English
• ISSNs: 1751-8687

Abstract

Abstract : Artificial Neural Network (ANN) has been recognized as a powerful method for short‐term load forecasting (STLF) of power systems. However, traditional ANNs are mostly trained by gradient‐based learning algorithms which usually suffer from excessive training and tuning burden as well as unsatisfactory generalization performance. Based on the ensemble learning strategy, this paper develops an ensemble model of a promising novel learning technology called extreme learning machine (ELM) for high‐quality STLF of Australian National Electricity Market (NEM). The model consists of a series of single ELMs. During the training, the ensemble model generalizes the randomness of single ELMs by selecting not only random input parameters but also random hidden nodes within a pre‐defined range. The forecast result is taken as the median value the single ELM outputs. Owing to the very fast training/tuning speed of ELM, the model can be efficiently updated to on‐line track the variation trend of the electricity load and maintain the accuracy. The developed model is tested with the NEM historical load data and its performance is compared with some state‐of‐the‐art learning algorithms. The results show that the training efficiency and the forecasting accuracy of the developed model are superior over the competitive algorithms.