Reservoir Water Level Prediction Using Supervised Intelligent Committee Machine Method, Case Study: Karaj Amirkabir Dam

The proper prediction of reservoirs water level variation is considered as one of the important issues for management, designing, operation of dams and water supply. In this study, based on five soft models such as support vector regression (SVR), adaptive neuro-fuzzy inference system (ANFIS), artificial neural network (ANN), radial basis function neural network (RBFNN), generalized regression neural network (GRNN) and the combined use of their results as input to one of these five models, a new structure called supervised intelligent committee machine (SICM) was proposed to predict the monthly reservoir water level of Karaj Amirkabir dam. The data used in this paper are water level, precipitation, evaporation, inflow and outflow of the dam. The evaluation of these models was done by nine error indexes and also the best model between them was selected using vikor decision maker method. After performing the necessary evaluations among the used soft models, the ANN known as the best model with nash–sutcliffe efficiency (NS) and mean square error (MSE) equal to 0.89 and 23.37 square meters, respectively. The results of the proposed approach are shown that the supervised (hybrid) neural network (SICM-ANN) has been able to provide high performance in predicting the monthly reservoir water level in Karaj dam with increasing the NS coefficient to 0.94 and decreasing the MSE index to 12.85 square meters (more than 45 percent decrease). Accordingly, hybrid use of soft models can be effectively applied to significantly reduce the predicted error of water level rather than single models.