Wind Speed Modeling based on Artificial Neural Networks for Jeju Area

This paper develops and evaluates a wind speed prediction model for Jeju area based on artificial neural networks, aiming at providing an accurate estimation of wind power generation to the smart grid system. For the history data accumulated for 10 years, the monthly speed change is modeled mainly to find the seasonal effect on tracing and resultant error patterns. A 3-layer model experimentally selects the number of hidden nodes to 10 and learns from 115 patterns, each of which consists of 5 consecutive speed values as input and one estimation output. The evaluation result shows that the error size is less than 5 % for 50 % of tracing and that slow charging over the median value opens a chance of further improvement. Finally, the monthly model makes it possible to build a refined day-by-day and hour-by-hour wind speed model based on the classification of months into winter, rainy, and other intervals.