Research On Methods And Tool Of Intelligent Simulation Optimization

In the design and decision in simulation models, the final purpose is tooptimize the performance measures based on output by adjesting the inputparameters of a simulation model. Simulation optimization is the process of improvethe search efficiency. It is very common for simulation models to have a lot ofvariables, multi-extremum, expensive running time, deicrete and continuousvariables coexistent and so on, which makes the optimization based on simulationmodel is more difficult compared with traditional optimization method.This papermainly discuss the simulation optimization method besed on metamodel, which isthe optimization throng metamodel. It solves the problem of limited simulationrunning times.Firstly, this paper give the process of the simulation optimization based onmetamodel. There are many types of metamodels and intelligent optimizationmethods, which have their advantages and disadvantages. This paper introduced themethod based on support vector machine and the particle swam optimization andgive the validation of the rationality. A flow can be designed to operate thesimulation optimization intelligently based on the selection strategy of commenmetamodel and research optization methods. From that, researchers can select theright metamodel and optmization method to receive expected results.The next step is to achieve an intelligent simulation optimization softwareusing metamodel. Before realizing the tool, there are two stages that are requirementanalysis and software design, which is expounded by use case and data flow diagram.It can provide strong guarantee for the use of the optimization flow and instanceanalysis futhur.Finally, for an example of simulation optimization simulation optimizationmethod and software are applied on. From the result of the test, the performance ofthe metamodel is satisfied. The predicted optimization value is better than theknwon one. Through the contrast test of the unitary and not unitary, and supportvector machine and neural networks, the correctness and practicability can bereceived.