| Summary: | 碩士 === 國立清華大學 === 工業工程與工程管理學系所 === 106 === This research is based on Particle Swarm Optimization (PSO) and combine with Ranking and Selection (R&S) method to solve the discrete event simulation optimization problem with single stochastic constraint. It needs to find the optimal or nearly optimal solution in finite time or budget when solving these problems. Because the constraints of problem exist stochastic, there are many feasible and infeasible solutions existing in solution space simultaneously. When dealing with such issues, researchers used heuristic algorithms to solve the problem of excessive number of solutions, and used R&S method to solve the sampling resource allocation problem in the past. However, there is still no method to consider the factors such as excessive number of plans, feasibility of plans, and sampling resource allocation simultaneously. This makes the problem with time-consuming in simulation and inefficient in solution solving process. Based on Optimal Simulation Budget Allocation for Constrained Optimization (OCBA-CO), this research proposed an Optimal Sample Allocation Strategy for Constrained Optimization (OSAS-CO) method. This method considers the variability and feasibility of solutions, and adds the concepts of Super individual and Elite group to allocate resources on key solutions for increasing the probability of selecting the best solution. Then applied to PSO to construct a hybrid algorithm.
According to the characteristics of PSO, the proposed method can improve the problem of excessive number of solutions that makes simulation time consuming. According to the characteristics of OSAS-CO, our method can also evaluate the feasibility of solutions while allocate the repetitive simulations. Then it improves the problem that PSO combines OCBA-CO will consume a lot of simulation computation budgets on solutions which have similar performance, and increase the sampling resource allocation efficiency. This research used the hybrid algorithm to solve the simulation optimization problem with single stochastic constraint.
This research used two different functional models and the buffer allocation problems respectively, and reduced the total simulation number by 57%, 14.4%, and 21.96% in three experiments respectively. This research proved that OSAS-CO can improve the usage efficiency of simulation computation budgets and reduce the total simulation numbers significantly.
|
|---|
