Abstract: In short-term optimal operation of a cascade hydropower system, the influence of water flow hysteresis between cascade reservoirs is usually neglected or the time delay considered as a constant. However, water flow hysteresis changes dynamically with upstream reservoir outflow, river channel storage, interval inflow and other factors. It is difficult to accurately predict downstream reservoir inflow in the cases of dynamic water flows. In this study, we screen out main factors affecting the inflow of lower reservoirs using a variety of methods, and develop a prediction model of the lower reservoir inflow by employing artificial neural networks with these factors as inputs. In a case study of the Jindong and Guandi hydropower stations, a daily optimization scheduling model of cascade reservoirs is constructed considering the dynamic time delay and solved with a progressive optimization algorithm. The results show that compared with the constant time delay models, a dynamic time delay model of a cascade reservoir system is more accurate in modeling the water flow connection between the reservoirs and can give an optimized scheme that increases the system’s power generation efficiency to a certain degree.

Key words: short-term optimal operation, water flow hysteresis, artificial neural network, cascade reservoirs, progressive optimization algorithm