Application of Combined Artificial Neural Network Model and meta-heuristic Optimization Algorithms in Predicting SPEI12 Drought Index

Drought is one of the most important and damaging natural disasters in the field of water resources that occurs in all climatic regimes of the country. Therefore, predicting and dealing with it is very important. In the present study, 79 synoptic stations in Iran were selected as the study. Three meta-heuristic optimization algorithms TLBO, IWO, PSO and the conventional Levenberg-Marquadt algorithm were used to train the multilayer artificial neural network to predict the SPEI12 drought index for the next one to three months. Due to the large number of synoptic stations, the stations were divided into five clusters C1 to C5 according to the time series of the drought using the K-means method. The results were compared with respect to the location of the stations in the clusters and the accuracy of the models was evaluated based on the RMSE and R2 indices of the test data. Showed that in all three prediction models, the accuracy of the models decreased with increasing prediction time. Comparison between the three optimization algorithms mentioned and Levenberg-Marquadt algorithm as a widely used algorithm in optimizing neural network weights, showed the better performance of meta-heuristic algorithms. The comparison between the three TLBO, IWO and PSO algorithms showed that the TLBO algorithm performed slightly better than the other algorithms and provided more accurate results. R2 was observed in cluster one (eastern regions, southern strip and southeastern regions of Iran) and the highest RMSE values and the lowest accuracy of the models were observed in cluster five (northern strip strip of the country).