Reservoir Outflow Prediction Using Adaptive Neuro-Fuzzy Interface System

Water supply, flood control, and hydropower generation are among the primary functions that rely on the prediction of reservoir outflow. Accurate prediction of reservoir outflow can help mitigate the flood risk and ensure the long-term sustainability of the hydraulic structure. Traditional prediction methods, such as linear regression analysis, and autoregressive integrated moving average (ARIMA) models, are based on historical data from the watershed. However, due to the uncertainty involved in the prediction of trends in flood events, the use of traditional methods is considered inadequate. To overcome these limitations, machine learning (ML) algorithms such as artificial neural networks (ANNs), support vector regression (SVR), and adaptive neuro-fuzzy inference systems (ANFIS) are used. In this study, ANN and ANFIS techniques are used for predicting the daily flow at the outlet of the Godavari river basin, located in West Godavari district in the province of Andhra Pradesh, India. Three key inputs, i.e. gauge data from the rain gauge station, discharge, and elevation data were used to develop the models. Input vectors for simulations included different combinations of antecedent precipitation and flows, with different time lags. From the results, it is observed that the outflow predictions obtained using the ANFIS model are more accurate as compared to those obtained from the ANN models. Lastly, a sensitivity analysis was performed to assess the reliability and resilience of both the neural network models.