Annual and Monthly Dam Inflow Prediction Using Bayesian Networks

Dam inflow prediction is important in terms of optimal water allocation and reduction of potential risks of floods and droughts. It is necessary to select a suitable model to reduce uncertainties in long-term and short-term predictions. In this study a probabilistic model of Bayesian Networks (BNs) was used to evaluate its efficiency in predicting inflow into reservoirs considering the uncertainties. For this purpose, continuous BNs as well as integration of K-means clustering and discrete BNs were applied for predicting magnitude and range of inflows, respectively in terms of annual and monthly prediction scenarios. In this regard, the Zayandehrud Dam reservoir in Iran was selected to test this model. To achieve the best network structure in these scenarios, different patterns were defined based on the combination of predictors. According to the magnitude predictions, the MAPE and R² indicators in annual model were respectively 21% and 0.62 and in monthly model were respectively 49% and 0.71. According to the results of the inflow range prediction, the prediction accuracy of the annual and monthly patterns was 75% and 83%, respectively. Modelling results showed that BN performs better in predicting the inflow range than its numerical prediction. The proposed model can improve the decision making of reservoirs operation.