Reservoir Inflow Forecasting using Artificial Neural Network

Accurate real-time forecasts of natural inflows to multipurpose reservoirs are required to provide basic information for reservoir management in a multipurpose water system. Multilayer perceptron (MLP) artificial neural network (ANN) models with Levenberg–Marquardt (L-M) algorithm were formulated based on different combinations of input parameters to predict one-day later inflow to reservoir. Best architecture of MLP ANN model to predict one-day later inflow to Hanuman Sagar reservoir at Wan, District Akola (India) was determined on the basis of statistical performance measures. The model no. 2 (architecture 4-8-1; inputs: one day ahead rainfall, current day rainfall, one day ahead inflow and current day inflow) predicted one-day later inflow excellently with 0.20 MSE, 0.93 r, 0.86 Nash– Sutcliffe coefficient of efficiency and 0.70 persistence index. The study confirmed that the evaporation and maximum temperature, as an input parameter, did not have any significant impact on prediction of inflow to the reservoir using ANN technique.