Performance Comparison of Hybrid Neuro-Fuzzy Models using Meta-Heuristic Algorithms for Short-Term Wind Speed Forecasting

Abstract

In this paper, short-term wind speed forecasting models have been developed using neuro-fuzzy systems. The optimal neuro-fuzzy model has been investigated in detail. In addition, meta-heuristic algorithms, such as artificial bee colony differential evolution genetic algorithm and particle swarm optimization for training adaptive neuro-fuzzy inference systems parameters have been used in this study. This is a novel study, as four different metaheuristic approaches are used to determine the appropriate adaptive neuro-fuzzy inference systems model parameters for short-term wind speed estimation, and analyzed comparatively. To validate the effectiveness of the proposed approach, wind speed series collected from a wind observation station located in Turkey are used in the short-term wind speed forecasting. In the first step, the results of analysis for finding the accurate model revealed that the optimal model that is proposed is adaptive neuro-fuzzy inference systems7-1 architecture. The meta-heuristic algorithms used in the optimization of adaptive neuro-fuzzy inference systems model parameters are then independently run 10 times, and the performance results are calculated statistically for the training and test phases of the adaptive neuro-fuzzy inference systems model. The results of the study clearly show that the adaptive neuro-fuzzy inference systems-particle swarm optimization hybrid model has the best performance in the training aspect, but it is observed that the ANFIS-differential evolution hybrid model gives better results than the others in the test step.