Efficient optimization procedures for stochastic simulation systems

In the research presented, we applied the particle swarm optimization (PSO) technique to optimize a concrete delivery operations simulation model (named HKCONSIM), aimed at improving the overall operational efficiency by minimizing the nonproductive time incurred on multiple building sites. Along with the conventional "steady, averaging" simulation-optimization mechanism, we proposed, assessed a "non-steady, stochastic" optimization mechanism, and further compared PSO with GA in applying the two mechanisms on a case study. It was found PSO was able to rapidly find the optimum on an output of a stochastic simulation model in the "non-steady, stochastic" setting, while GA failed to converge. Compared with the performance of GA on the conventional "steady, averaging" optimization setting, the use of PSO in the "non-steady, stochastic" setting resulted in a marked improvement in light of the optimum-seeking time, requiring about 5 minutes PC time as opposed to about 1.5 hours taken by GA. Therefore, the proposed PSO-based, "non-steady, stochastic" optimization procedures can comfortably, rapidly approach the optimum state for a large-scale, complex system simulation of realistic granularity.