ABSTRACT
The direct optimization methods based on operation rule curve generalization have the primary problem with a very narrow or irregular feasible solution space. Although the existing optimization algorithms have a better self-learning ability, satisfactory optimal solution cannot be acquired in a short time, constrained by computational capacity of common computers. At present, there are no ideal solutions to this problem, either
1 INTRODUCTION
As a simple and practical implicit stochastic operation method, reservoir operation rule curves have been widely applied in practical operation. Operation rule curves can be regarded as a set of the envelope curves meeting the system security to conform to design dependability in a selected hydrological year. The conventional operation rule curves are often thought to reflect history cases, the reliability of the normal conditions of the reservoir and the relative rationality of water use. In view of that, it becomes difficult to reach the global optimum and quasi-global optimum, and deal with the complex issues of multiple objectives.
