Modeling the Monthly Inflow to Jamishan Dam Reservoir Using Autoregressive Integrated Moving Average and Adaptive Neuro- Fuzzy Inference System Models

Hydrological time series modeling is one of the most important issues in water resource management. In this paper monthly inflow to Jamishan dam reservoir in Kermanshah province (west of Iran) is modeled by AutoRegressive Integrated Moving Average (ARIMA) and Adaptive Neuro-Fuzzy Inference System (ANFIS) models. These models are based on stochastic and Artificial Intelligence (AI) methods, respectively. For modeling up to five parameters in the ARIMA model were used and produced 1296 models which were fitted on the time series. In ANFIS model 14 input combinations were defined using the discharges with different lags. Two states of Grid Partitioning (GP) and Subtractive Clustering (SC) were used in Fuzzy Interface System (FIS) generation. Also, in training network Back Propagation (BP) and hybrid algorithms were used. Monthly modeled discharges were compared in the ARIMA and ANFIS models by some indexes such as Mean Absolute Relative Error (MARE) index which was obtained 0.398 and 0.8 for each model, respectively. The result showed that the ARIMA model was much more accurate than ANFIS model in modeling low discharges and also in short and long times modeling.