Application of Time Series, ANN, and SVM Models in Forecasting the Gorgan Dam Inflow Rate

In water resources management, there is a critical need to the prediction of the amount of inflow into the water supply system in order to be aware of future conditions and planning for optimal allocation of water resources to different sectors such as drinking, agriculture and. The aim of this study is to forecasting the monthly inflow to the Gorgan dam for future. To this aim, the data of the Qazaghli station with a 47-years history period and three Time series, neural network and Support vector machine models used for prediction. According to the obtained results, the ARIMA (1, 0, 0) (1, 0, 1) was found to be the premier parsimonious time series model based on the Akaike and Schwarz criteria. Moreover, The ANN model with 2 input and 10 neurons tuning and the SVM model with one input were the best performing models. Finally, according to the obtained results and evaluation criteria, the SVM model has the best efficacy in comparison with two other methods. The RMSE and AARE was 5.31 and 1.07 for SVM model, respectively; 9.88 and 2.78 for neural network, respectively and 8.84 and 1.07 has been obtained for Time Series model, respectively. Based on the results of this research, the best model to predict the monthly discharge input to the Gorgan dam was SVM.