SIMULATING AN EVOLUTION STRATEGY TO FORECAST TIME SERIES ARMA(1,1) MODEL

Abstract

This paper presents a multi-member evolution strategy to forecast future value of observed time series model. The proposed method is simple and straight forward and doesn't required any problem specific parameter tuning of the problem. The experiments designed based on simulate for different values of sample size (n=25,50,100),model parameters set ( ) and set to ( ) and use lead time for forecasting future value equal to (l=1,2,3).The value of take equal to (15,100) beside this, there is anther experiment designed for simulating one of method which is known as Box –Jenkins with same values of sample size, model parameters and leads time(l) for number of replicate (RR=1000). Results of this study has cleared by numbers of figures and tables, which are made to clear compression between ES-algorithm and B.J method based on computing values of FMSE (Forecasting Mean Square Error ) & Thiels' (U- statistic) ,statistics used as tools to measures reliability of ES- algorithm and also used to clear accuracy of ES algorithm results. Table(1), tables (2-7) and figures (4 -9) results of statistics show the reliability of algorithm to producing individuals which give reasonably predictions of future values of time series for different values of sample size and lead time values of model parameters.