Comparison of different models for wind speed prediction

Abstract

The paper presents several unconventional models for wind speed prediction based on fuzzy logic and neural network techniques. First, two fuzzy models of a position and position-gradient type are built on the basis of different meteorological data such as solar radiation, relative humidity, ambient temperature, atmospheric pressure etc. In order to obtain the fuzzy models for wind speed prediction, Sugeno-Yasukawa identification algorithm was employed. Next, a neuro-fuzzy model for wind speed prediction was build, based on statistical learning theory. The model presents a fuzzy inference system of Takagi-Sugeno type that uses an extended relevance vector machine for learning its parameters and number of fuzzy rules. Finally, a neural network approach was applied to build two different models for wind speed prediction based on extreme learning machine techniques. Both neural models represent single layer feedforward neural networks, with different learning algorithms. The first one applies classic extreme learning machine and the second one uses incremental extreme learning machine philosophy. The obtained models are compared for their generalization performance and approximation capability, and although they all possess excellent approximation capabilities, the neural model based on incremental extreme learning machine has shown the best simulation results.