Evaluation of the combination of optimization algorithms and adaptive fuzzy-neural inference system compared to time series models in groundwater level estimation

To optimize the management and optimal use of groundwater resources, it is necessary to be aware of the temporal-spatial changes of the stagnant level . For modeling and predicting hydrogeological variables, the question remains:To what extent these hybrid models can be effective compared to the individual model?, in this study four algorithms of particle overvoltage optimization (PSO) genetics (GA) ant colony (ACOR) and demand evolution (DE) were combined with the model of adaptive fuzzy-neural inference system (ANFIS).The four combined models performance developed with the ANFIS model and the time series model (SARIMA) as the reference model to estimate the average monthly groundwater level of the Sahneh plain aquifer in Kermanshah province was evaluated over 19 years.To better compare the results of the models, the same input variables of the groundwater level in different time steps (maximum four months based on the self-correlation function of aquifer level) were considered for them. The results of fitness indicators in the test and test phase showed that there was no significant difference between the SARIMA time series model compared to other combined models used.However, given that SARIMA applies average moving processes, authorization, seasonal changes, and delays in modeling, groundwater leveling can be given more attention in modeling. The RMSE values of the best hybrid model (ANFIS-GA) and SARIMA were 0.950 and 0.1012, respectively. The results also showed that the combination of optimization algorithms considered with the ANFIS model does not improve the model's results compared to the individual ANFIS model in terms of significance.