Forecasting of Inflow to Kamal Saleh Reservoir using Soft Computing

Forecasting of inflow to reservoirs is essential in efficient planning and operation of water resources. In this research¡ the application of two different intelligent models including Artificial Neural Networks (ANNs) and Adaptive Neuro-fuzzy Inference System (ANFIS) technique based on subtractive clustering method are investigated for forecasting inflow to Kamal Saleh reservoir in Markazi province¡ Iran. For this purpose precipitation and discharge data in a 31-year statistic period (1981-2011) were used and forecasting of inflow was done in the daily and monthly time steps. The amount of discharge and precipitation in previous time steps were used as input patterns for models.Both ANNs and ANFIS models had very acceptable performances at daily and monthly forecasts of inflow based on the error measures; R¡ RMSE and MAE¡ however ANFIS model performance was better than ANNs (less than 3%). Using seasonality coefficient (α) improved the performance of models in monthly forecasts. In the following¡ the effect of large scale climate variables including North Atlantic Oscillation (NAO) index and Southern Oscillation Index (SOI) examined on monthly forecast that was resulted from the best input pattern in optimized model of the previous section. The results showed that using climatic indices in input combination can improve the performance of model to forecast the inflow. Although SOI had greater impact on improving the monthly inflow forecast. So that the statistical indices to evaluate the error of ANFIS model including R¡ RMSE and MAE obtained as 0.91¡ 3.56 and 3.73 respectively¡ this shows the potential of this way to increase the precision by improving the evaluation criteria about 11%¡ 9% and 11%¡ respectively.