Comparison of Techniques and Input Data for Neural Networks in Prediction of Wind Speed
Abstract
In recent years, renewable energy sources have gained importance worldwide and became one of the most important sources to supply the electrical energy demand. Wind energy have been highlighted as one of the cleanest technologies, showing wide applicability. In the Brazilian state of Ceará there was in the last years an investment of about 644 million euros in the construction of 250 towers for wind energy production, which are distributed in 14 parks. The total electrical output of these wind parks is 500 MW. Ceará is the Brazilian state with the highest installed wind power but this represents only 5% of the state wind potential. The installation and operation of a wind farm needs a planning that requires forecasts of wind speeds in order to get an estimation of the plant generation capacity. Application of Artificial Neural Networks (ANN) to forecast wind has proved a convenient and efficient technique. In this way, the present paper has as main goal the use of ANN to make wind speed forecasts based on meteorological data measurements as input. Additionally, the paper compares the results of applying different ANN methodologies available to estimate the wind potential, as well as the effect of different input data and the combination of the data (wind speed, ambient temperature, air humidity and time of the measurement) to forecast the wind speed.The remainder of the paper is organized having s2 review, section 3 presenting the ANN models, section 4 the methodology, section 5 the training results, ending with the conclusion in section 6.