Application of Genetic Algorithm in Estimation of Linear Time Series Parameters for the Purpose of Drought Prediction

So far, the linear time series parameters are estimated, generally based on graphical and approximate methods. Therefore, the use of a new approach to increase the speed and ease of access to the best time series model can play an important role in using this method for predicting hydrological events. In this research, an optimization approach based on genetic algorithm has been used to estimate the ARMA time series parameters. A hybrid of Genetic Algorithm-ARMA method was used to drought prediction at three selected stations in the Urmia Lake basin, including Tabriz, Saqhez and Urmia, based on the SPEI drought index. The results showed that according to the BDS test, the model had the ability to predict the drought in all three stations and in all time scales. The Ljung-Box statistic was also used to evaluate the reliability of the prediction model. Its p-value at all stations and time-scales were greater than 0.05 which indicated the residuals of models were random and reliable. Also, the best time series model was calculated at different time scales and based on this, the SPEI index was predicted. The results of the prediction section showed that ARMA-GA hybrid method had a high accuracy at all long-term time scales of SPEI index at all the stations, but its performance was not suitable for short-term time scales.