Echo state networks, artificial neural networks and fuzzy systems models for improve short-term wind speed forecasting

This article presents results from models developed to improve wind speed forecasting based on techniques of echo state networks, multi-layer perceptron (MLP) and neuro-fuzzy system (ANFIS). Wind generation is a direct function of wind speed and, in contrast with conventional generation systems, is not easily dispatchable. Wind forecasting is crucial both for wind farm operators and utility operators. Accurate forecasting allows operators to achieve favorable trading performances on the electricity markets. The further in advance an operator can make a reliable estimate about how much electricity he will produce the better. However, the use of wind to generate electricity has some drawbacks, such as uncertainties in generation and some difficulty in planning and operation of the power system. Models presented in this paper investigated the contribution of using climate variables such as humidity, temperature and radiation, in order to improve the performance of wind speed forecasting. The models were adjusted to forecast the wind speed with steps up to four hours ahead. The inclusion of climate variables have enabled significant gains in wind speed forecasting results. The gain of some models developed, in relation to the reference models, were approximately 50% for forecasts in a period of four hours in advance.