半Markov控制过程基于性能势仿真的并行优化算法

根据等价Markov过程方法,研究了一类半Markov控制过程在紧致行动集上关于无限水平平均代价准则的性能优化算法.由于实际系统的状态空间往往非常大,因此通常的串行仿真算法可能会耗时过长,或由于硬件限制而无法实现.针对这些问题,提出了一种基于性能势的并行仿真优化算法,以期寻找系统的最优平稳策略,并用该算法对性能势的仿真和策略寻优分别进行了并行化,获得了较好的运行效率.仿真实例表明了该算法的有效性.这一算法可应用于大规模实际半Markov系统的性能优化.

Parallel optimization algorithms for semi-markov control processes based on performance potentials simulation

Based on the equivalent Markov process,performance optimization algorithms were studied for a class of semi-Markov control processes (SMCPs) with infinite horizon average-cost criteria and compact action set.Since the state space of a practical system is often very large,when applying traditional serial simulation algorithms,a long time is possibly required,and it is impossible to realize the algorithm due to limitations of the hardware.A parallel simulation optimization algorithm based on performance potentials was proposed to find the optimal stationary policy of a system.In this algorithm,the simulation of the performance potentials and the part of policy iteration are paralleled respectively,and high efficiency was achieved.A simulation example shows that the algorithm can get high speedup.The algorithm can be used in optimization for large-scale practical semi-Markov systems.