Application of Genetic Algorithm to Optimize the Performance of Adaptive Neural - Fuzzy Inference System in order to predict maximum of air temperature (Case study: Isfahan city)

In this study, the use of genetic optimization algorithm (GA),Particle Swarm(PSO), the ant colony for continuum (ACOR)and differential evolution (DE),to develop and improve the performance of ANFIS were investigated. the monthly maximum temperatures in Isfahan during the period of 64 years (1951-2014), was simulated and analyzed. At first in a sensitivity analysis, the best entries for each prediction period (1 month, 1, 2 and 3 years) were selected. Then, the maximum temperature hybrid models by ANFIS-GA,ANFIS-PSO,ANFIS-DE,ANFIS-ACOR and ANFIS were examined. The performance of each model with regard to R2, RMSE and MAE were evaluated. The results showed that the ANFIS-GA, as the most appropriate model, increased ANFIS performance in R2 to by 0.06, 0.07, 0.08 and 0.12 and RMSE by 0.09, 0.09, 0.16 and 0.1, respectively, in 1 month and 1, 2 and 3 year. After, ANFIS-DE and ANFIS-PSO, respectively, had the best forecasting accuracy. On the other hand, ANFIS showed highest error and lowest R2, as the weakest model. The results showed that the proposed models, which use global search techniques and avoid being trapped in local optimum, could improve the performance of ANFIS favorably.Therefore, these models can be used in other areas related to hydrology and water resources.