Assessment of Stochastic Models and a Hybrid Artificial Neural Network-Genetic Algorithm Method in Forecasting Monthly Reservoir Inflow

Abstract

In this study, the performance of Seasonal Autoregressive Integrated Moving Average (SARIMA) models and hybrid Artificial Neural Network-Genetic Algorithm (ANN-GA) method in forecasting the monthly inflow to a dam is examined and compared. The number of parameters required for SARIMA models is determined using Autocorrelation Function (ACF) and Partial Autocorrelation Function (PACF) diagrams. In terms of these diagrams, 14 input combinations are considered for the ANN-GA models. An examination of the model’s performance in prediction utilizing ANN-GA indicates that applying discharges with 1, 2, 6, and 12-month lag time leads to the best prediction. The model results are compared using mean absolute relative error (MARE) and correlation coefficient (R) indexes. These index values for the SARIMA model are 0.388 and 0.76 and for the ANN-GA model they are 0.815 and 0.8, respectively. Moreover, to evaluate the ability of the models to make short-term and long-term predictions, relative error (Ei) and average of the cumulative relative error (Fi) are computed for the forecasted discharges. The results indicate that the SARIMA model is more capable in forecasting monthly inflow, especially for low values, than the ANN-GA model. The SARIMA model is also much more accurate than the ANN-GA model in short-term and long-term forecasting.