Research On Sensitivity Analysis Methods And Tool Based On Simulation Experiments

Compared with the traditional experiments, simulation experiments have morefactors, more levels related to factor, and more indexes. It’s impractical to usesensitivity analysis qualitatively analyzing each factor, which uncertainty influencedthe uncertainty in the output of a model. And the cost of experiments is expensivetoo. The Pareto theorem or20-80rule reckons that only a small amount of factorshave important effects on the simulation output. To solve this problem, this papercarried out as following several aspects, combining with the characteristics ofsimulation experiments and the perspective of sensitivity analysis.Firstly, experimental analysts should screen out important factors, andquantitatively analyze the sensitivity coefficients of these factors according to thecharacteristics of the simulation experiments. Based on this idea, this paperproposed the process model of sensitivity analysis, studied and compared globalsensitivity analysis methods. According to the characteristics of the experimentaldesign method and sensitivity analysis, intelligent technology which based onhybrid reasoning was proposed. This technology has the role to guide experimentalanalysts fulfilling the work reasonably and efficiently.Secondly, due to the factor screening and quantitative analysis the core of thesensitivity anslysis, only in the understanding of screening methods canexperimental analysts screen factors efficiently. Meantime, sensitivity analysisqualtitatively or quantitatively need sample data from sampling. Therefore, severalkinds of screening methods including Sequential Bifurcation and FractionalFactorial Design, and sampling methods including Monte Carlo Method and LatinHypercube Sampling were studied in this paper.Finally, the process of sensitivity analysis is quite complicated and heavyworkload, which requires exprimental analysts with certain knowledge reserves. Inorder to improve the efficiency and accuracy of the sensitivity analysis, this paperdesigned and fulfilled a sensitivity analysis tool, assisting exprimental analysts toanalyse sensitivity. The tool is designed with phased projcet management model andbased on hybrid reasoning model, and applied to an actual problem of sensitivityanalysis effectively.