부산대학교 도서관

A generic and scalable scheme is proposed for forecasting reservoir inflow to optimize reservoir operations for hydropower maximization. Short-term weather forecasts and antecedent hydrological variables were inputs to a three-layered hydrologically-relevant Artificial Neural Network (ANN) to forecast inflow for 7-days of lead-time. Application of the scheme was demonstrated over 23 dams in U.S. with varying hydrological characteristics and climate regimes. Probabilistic forecast was also explored by feeding ANN with ensembles of weather forecast fields. Results suggest forecasting skill improves with decreasing coefficient of variation in inflow and increasing drainage area. Forecast-informed operations were simulated using a rolling horizon scheme and assessed against benchmark control rules. Over two years of operations from Pensacola dam (Oklahoma), additional 47,253 MWh of energy could have been harvested without compromising flood risk with optimal operations. This study reinforces the potential of a numerically efficient and skillful reservoir inflow forecasting scheme to address water-energy security challenges. • ANN reservoir inflow forecasting scheme is proposed based on weather forecast. • The data-based forecasting scheme is efficient and scalable. • Tests reveal robust performance across diverse hydro-climatic regimes. • The scheme can be used for reservoir operations for hydropower maximization. [ABSTRACT FROM AUTHOR]