演化仿真优化的若干问题研究

The Method and Application of Evolutionary Simulation Optimization

针对仿真优化这一仿真领域提出了演化仿真优化的概念及其形式化语言描述 ,并给出了演化仿真优化的算法实现框架、分类 ,指出了仿真优化与演化算法相互取长补短的策略 .为了提高速度和效率提出了一种混合演化仿真优化的算法 :基于拉网的仿真优化算法 (MESOAs) .该算法结合仿真系统的输出信息 ,构造出系统的响应曲面来指导演化算法 ,同时又不要求系统连续、可导 ,因此具有通用性、鲁棒性、隐含并行性等优点 ,它能有效地解决不确定环境 (含随机系统和定性系统 )的仿真决策优化 ,连续时间仿真控制优化问题等 .最后 ,给出了两类测试演化仿真优化算法的测试实例 :一类是随机函数 ,另一类是 GI/ G/ S排队模型 ,试验的结果表明在解的质量和速度两方面 MESOAs都优于曲面响应法。

Evolutionary computation, which is wildly applied to variant fields, is an effective stochastic search method. This paper proposes the concept of Evolve Simulation Optimization (ESO) and its formal language description. The realized framework and classification of ESOs are also proposed. Because on the one hand the simulation optimization process consumes lots of CPU time, and on the other hand the evolutionary algorithms lack of guideline of simulation system information, if we only combine simulation optimization and evolutionary algorithms, the performance of ESO algorithms isn't better than the other SO algorithms. So author designed a hybrid Evolve Algorithms, which called Mesh\|based ESO algorithms (MESOAs), to enhance the speed and efficiency of ESO algorithms. The MESOAs utilize the outputs of simulation system to construct system responding surface and then gain the gradient information to guideline the Evolve Algorithms. At the same time the MESOAs doesn't require that the simulation systems are successive and derivative, So general, inherited parallel, robust and global search are the advantages of MESOAs. This paper also gives some application instances in the field of Simulation Optimization (SO), which include stochastic system, discrete GI/G/S queue models, and so on. The experiment results show that the MESOAs can effectively solve the problem of simulation decision\|making optimization of indeterminated environment, control optimization design of continue time simulation system.