126. A NEURO-FUZZY MODEL FOR WIND SPEED PREDICTION BASED ON STATISTICAL LEARNING THEORY
Keywords: wind speed prediction, neuro-fuzzy modelling, extended relevance vector machine, kernel function, relevance vectors
AbstractWind is free, clean, and renewable source of energy and is fast becoming a desired alternative to conventional energy resources such as fossil fuels. That is why more and more countries are intensifying their efforts in wind energy research and harnessing. Among other wind characteristics, wind speed is crucial for planning, de¬signing and operating wind energy systems. This is the reason for much research in the field of wind speed modelling and prediction. There are many research papers dealing with the problem of forecasting the wind speed, which re¬quires special attention because of time-varying, stochastic and intermittent nature of wind. It has been shown in lit¬erature that among the many proposed models for wind speed prediction, the models based on soft computing tech¬niques such as artificial neural networks, neuro-fuzzy inference systems and machine learning are superior in terms of approximation accuracy. While there are many neural models for wind speed prediction that deploy different learning methods, and there are many hybrid models based on fuzzy logic, neural networks and genetic algorithms etc., the re¬search conducted in this work has shown that practically there are no neural models based on relevance vector ma¬chine and no neuro-fuzzy models that apply RVM learning mechanism, which is state of the art technique. This paper presents possibly for the first time in literature a neuro-fuzzy model for wind speed prediction based on Vapnik’s sta¬tistical learning theory, Tipping’s relevance vector machine and Kim’s fuzzy inference system. The model is a fuzzy inference system of a Tagaki-Sugeno type that relies on extended relevance vector machine for learning its parame¬ters and fuzzy rules. The wind speed is modeled by means of available meteorological data such as total solar radia-tion, ambient temperature, humidity, atmospheric pressure, etc. The performance of the model is validated through its performance index and compared to other fuzzy and neural models for wind speed prediction. The simulation results show clearly that the model possesses excellent features and the best performance in terms of accuracy.