中文摘要：

随机动态规划（SDP）方法是水库优化调度的基本方法，但将其应用于包含有多年调节水库的水库群的优化调度时会引起“维数灾”问题，并且难以反映多年调节水库调节周期不定的特点。针对这种情况，本文提出多层次的改进遗传模拟退火优化算法（IGA—SA），把库群优化问题分解为第1层次的SDP优化与第2层次的IGA—SA优化，从而获得库群的优化调度结果，并应用于贵州乌江梯级水库群中长期发电优化调度研究中，取得较好的结果。实践表明，该方法可以克服随机动态规划应用中遇到的“维数灾”问题，并给包含有多年调节水库的水库群的优化调度问题研究提供了有效的工具。

英文摘要：

Stochastic Dynamic Programming（SDP） is a basic method for optimal operation of reservoirs. However, when the SDP is applied to the optimal operation of multi-reservoirs system with multi-annual regulating reservoirs, the problems of ＂curse of dimensionality＂ will be aroused and the characteristic of uncertain regulated cycles can＇t be showed. Aiming at these problems, a multistage optimization method of SDP with an improved genetic algorithms ＆ simulated annealing algorithms （IGA-SA） is presented in this paper. This method decompounds the optimal reservoirs problems of cascade hydropower stations into the first stage optimization of SDP and the second stage optimization of IGA-SA. The optimal operation results of cascade reservoirs are then obtained. The application of the long-middle term optimal reservoirs operation of cascade hydropower stations on Wujiang River demonstrates that the model operates very well, and is capable of overcoming the ＂curse of dimensionality＂ problem, it provides an effective tool and method for optimal operation problems of multi-reservoirs system.