Development of wind speed prediction model in Jeju City

Abstract

This paper develops and evaluates wind speed prediction models for Jeju City based on artificial neural networks, aiming at more integrating renewable energies into the power system. 3-layer neural network models take the appropriate training pattern from the history data accumulated during the last 10 years. First, the monthly model classifies the months into rainy, winter, and remaining periods according to the error size. The auto correlation function analysis confirms that the modeling error can be considered as white noise. Next, a 5-day forecast model takes wind speed for 5 previous days as inputs and generates 5 outputs for next 5 days. The 1-day advance tracing error is 1.28 mps (meter per seconds) in March and 0.66 mps in August on average. In addition, the prediction error is 0.45 mps for the next first day forecast and 1.99 mps for the 5-th day forecast in July. © 2012 Springer-Verlag Berlin Heidelberg.