Author(s): Kayhan Gavahi; S. Jamshid Mousavi; Kumaraswamy Ponnambalam
Linked Author(s):
Keywords: Daptive Reservoir Operation; Inflow Forecasting; KNN; ANFIS
Abstract: This study investigates the effect of two data-driven inflow prediction methods on the performance of a proposed adaptive real-time optimum reservoir operation model. The model consists of three modules; a forecasting module, which predicts the monthly future inflows, a reservoir operation optimization module, determining monthly optimum reservoir releases up to the end of a year, and an updating module, updating the current state of the system and provides the other two modules with the latest observed information on future inflows. K-nearest neighbour (KNN) and adaptive neuro-fuzzy inference system (ANFIS) approaches are used to forecast monthly inflows to the reservoir. The results demonstrate that ANFIS outperforms the KNN approach by 25,23,27 and 10 percent with respect to RMSE, PWRMSE, NSCE and correlation coefficient indices, respectively. However, the objective function values of the reservoir operation optimization model associated with each of those forecast models reveal that ANFIS-based adaptive reservoir operation model is only 5% better than the KNN-based model. This observation highlights the significance role of adaptation and updating procedure in the reduction of streamflow forecast errors.
Year: 2018