Intelligent self-developing and self-adaptive electric load forecaster based on type-2 fuzzy Bayesian Ying-Yang learning algorithm

Abstract

In new advocated "smart grid" development, an electric load forecaster should possess high-level intelligence in order to handle higher uncertainty, indefiniteness, and variability on electric load demand. The intelligence is referred to as self-learning, self-adaptability, and the highest capability of handling various uncertainties, which the forecaster should possess. In this study, a novel methodology, self-developing and self-adaptive fuzzy neural networks using type-2 fuzzy Bayesian Ying-Yang learning algorithm (SDSA-FNN-T2BYYL) is proposed. Its novelty is that (1) the Bayesian Ying-Yang learning algorithm (BYYL) is used to construct a compact system structure automatically. (2) Further, a novel T2 fuzzy BYYL is presented, which integrates type-2 (T2) fuzzy theory and BYYL in order to achieve two objectives simultaneously: compact system structure and better handling of data uncertainty. (3) Because a training dataset cannot include all possible operation conditions, the system should be able to restructure continuously for good generalization. Consequently, a T2 fuzzy BYY split-and-merge algorithm is proposed. The proposed method is validated using a real operational dataset collected from a Macao electric utility. Simulation and test results reveal that SDSA-FNN-T2BYYL has superior accuracy for load forecasting over other existing relevant techniques. © 2013 Copyright Taylor and Francis Group, LLC.