Efficiency Comparison of Bayesian and MLP Neural Networks in Predicting Runoff to the Taleghan Dam

The importance of regulating the supply and demand regime shows the need for planning in the exploitation of surface water resources. The aim of this study was to compare the performance of two models of Bayesian network (BN) with a probabilistic approach and MLP neural network for flow prediction and selection of the best structural model. Monthly meteorological data including rainfall, monthly average temperature, evaporation, and the volume of water transferred from five hydrometric stations were introduced as input data to the models, and runoff to the dam was considered as predictable. Input data with different layouts were introduced to BN and MLP models. The results were obtained by comparing 17 selected models according to the index criteria: Nash-Sutcliffe coefficient (NS), mean square error (MSE), mean square error root (RMSE), and MEAN absolute prediction error (MAPE). The best model in BN model with 43.3% similarity and index criteria was estimated to be -3.98, 300, 17.3, and 0.06, respectively. The MLP model with 80% similarity and index criteria were introduced as -10.3, -8266, 23.9, and 122.3 in the best model, respectively. As a result, both models performed well in runoff estimation, but the BN model had much better accuracy in forecasting. Finally, a structural pattern with acceptable results in both MLP and BN models was identified.