Cleavage self : a new concept in reproduction stage of genetic algorithm for rainfall prediction

Abstract

Rainfall prediction which considering climate variables such as air temperature, air humidity, air pressure, and wind speed are categorized to a non-stationary stochastic processes. Modeling of time series data of precipitation is carried out using MISO ARX model. Genetic Algorithm (GA) is used to optimize the entire model coefficients so that the results obtained is quite accurate. A new concept in the reproductive stage called Cleavage Self (CS) is designed to improve performance of GA in terms of speed and accuracy of the prediction process. In order to prove its performance, prediction of time series precipitation is also conducted by applying AG without CS. The result is the speed of the process AG with CS is approximately 9.6 times faster than that without CS. The absolute differences of the MSE ideal for AG with CS is 0.0004 compare to 0.0151 for AG without CS.