Complex Simulation-oriented Assessment And Optimization Methods

Simulation assessment involves assessment of simulation system or simulation model and simulation-based assessment.With the complexity of simulation objectives and the increase of simulation application requirements,some characteristics such as large-scale,non-linear,uncertainty,continuous discrete mixing,complex outputs and large amounts of simulation data etc.are raised in simulation system,leading to great challenges in the existing simulation assessment,optimization methods and tools.The research works focus on credibility assessment of complex simulation models,simulation-based effectiveness assessment and optimization for system-of-system(SOS).Simulation complexity is analyzed from the internal and external aspects,and the mathematical formulation of assessment and optimization for complex simulation are given.For credibility assessment of complex simulation model composed of multiple submodels when observation data of large-scale systems are lacking,a structural credibility extrapolation model based on the relationships between output variables is constructed.Bayesian networks are used to estimate the posterior distributions of system level outputs based on the submodels validation data,and the highest posterior density(HPD)confidence region-based criterion is proposed to validate the submodels and the system-level model respectively.In addition,the proposed validation criterion can effectively improve the accuracy of validation results when the observation data are sufficient.For non-stationary,multivariate,static and dynamic complex simulation outputs with correlation and uncertainty,a Hilbert-Huang transformation-based ranking assessment method for simulation models,and a result validation method combined variables correlation analysis with area metric are proposed.Non-stationary time series are decomposed into trend and stationary terms with empirical mode decomposition.A consistency analysis method for stationary terms is proposed based on Hilbert spectral density.The grey relational analysis is used to calculate the ranking result of simulation models.For multivariate,static and dynamic,correlated output variables,the mutual information and fractal dimension methods are applied to extract the correlated variable subsets of the dynamic and static outputs,respectively,and the simulation model credibility is measured based on the joint area metric under uncertainty.For the complex effect relationships among various equipments,and the requirements of efficient effectiveness simulation optimization for SOS,a multi-equipment coordination-oriented effectiveness assessment method and an effectiveness optimization method combined sequential metamodel with improved multi-objective optimization are proposed respectively.The association rule mining is applied to analyze the massive simulation data,and construct the networked assessment index system based on the index association relationships;the analytic network process is used for the effectiveness indexes integration.Multi-objective optimization algorithm NSGAII is improved to solve the continuous and discrete hybrid optimization problem,and combined with the metamodel method to enhance the efficiency of expensive simulation optimization.Then,an exploitation-exploration-based sequential strategy combining space filling and clustering is proposed to improve the metamodel accuracy.A Hadoop-based comprehensive assessment system for complex simulation(HIT-HCAS)is developed to satisfy the requirements of management and analysis for simulation big data,assessment and optimization for complex simulation.HITHCAS can be used to support the functions of distributed data management and parallel analysis,complex simulation results validation,credibility extrapolation for complex simulation models,simulation-based effectiveness assessment and optimization for SOS.Hadoop is applied to provide the environment for simulation data management and algorithms operation,and some characteristics such as scalable algorithms,reusable components,and process visualization are involved in the tool.The validity and practicability of the system are illustrated by the applications including simulation credibility assessment,effectiveness simulation assessment and optimization of an air defense SOS.Finally,the main innovations of this paper are summarized,and future research work is prospected.