A combination of an accurate reservoir inflow forecasting model with an efficient method of numerical optimization can be used to generate efficient and balanced solutions for operating multi-purpose reservoir systems and improve hydropower generation. In this study, a hybrid between a data driven artificial neural network (ANN) model and a novel combined Pareto multi-objective differential evolution (CPMDE) is presented. The methodology is applied for real-time hydropower production from the Vanderkloof in South Africa. ANN was employed to forecast daily reservoir inflows while CPMDE was used to generate optimal policies for the real-time operation of the reservoir for power generation. The objectives were formulated to maximize power production and minimize storage depletion. Results from the application of the real-time model indicate that energy may be generated from the reservoir throughout the year without system failure under normal flow conditions. The real-time method generates optimal policies that efficient lytrade-off the objectives by finding a balance between immediate and the future production of hydropower. This suggests that adopting real-time optimization strategies may be beneficial to operation of reservoirs.