Performance of different hybrid algorithms for prediction of wind speed behavior

Abstract

This study seeks to provide a new method by proposing three hybrid algorithms. The proposed algorithms include genetic neural network hybrid algorithm, simulated annealing neural network hybrid algorithm, and shuffled frog-leaping neural network hybrid algorithm. The efficiency and reliability of the presented hybrid algorithms in prediction of wind speed behavior were evaluated using meteorological data of the city of Abadeh for a 10-year period from 2005 to 2015. The forecasting horizon is monthly for this study. The study parameters consisted of TMAX, TMIN, VP, RHMIN, RHMAX, WIND SPEED, PRECIPITATION, and SUNSHINE HOURS. These eight parameters are used as the inputs, and one parameter (ET) is used as the output. Research findings show that the shuffled frog-leaping neural network hybrid algorithm providing a root mean square error value of 0.0761 and reliability of 0.91 is more suitable than other hybrid algorithms for prediction of wind speed behavior in the study area.