Abstract: Owing to uncertainties of hydrological forecast and directives, the operation rules play a very important role in managing reservoirs, and are the most commonly used and effective tools for the reservoir dispatching operation of hydropower stations, though the real time optimal operation based on mid- and long-term hydrological forecasting information which has been studied for many years. So the development of the reservoir optimal operation model for formulating operation rules has always been a research hotspot in the relevant field. A model named long-term reservoir optimal operation model for hydropower, based on genetic algorithm and stochastic dynamic programming (hereafter referred to as LROOH) is established, which couples the stochastic dynamic programming and the real coding accelerating genetic algorithm. This model solves the difficult problem with a satisfied scheme, via building the objective function and minimizing the absolute values of the difference between the calculation reliability of the guaranteed capability and its target reliability, with penalty coefficient as an independent variable, and making full use of the parallel computing ability of genetic algorithm. And then the LROOH becomes much easier to have access to the global optimal solution by using the real coding accelerating genetic algorithm instead of 0.618 methods that are usually used before to improve the computing efficiency. As an example, the LROOH is applied to an annual regulation of a reservoir of the hydropower station. The research results show that the model is effective, with some excellent properties that are without any constraint of annual and multi-year regulating storage, and that the optimal rules can meet the requirements of the guaranteed reliability and increase output by more than 2.0% with a higher reliability.