Study On Fitting Methods Of Simulation Metamodelling And Its Application

Simulation is ranked as a powerful tool for complex system modeling, which is widely used in performance evaluation, system design and decision support. The simulation for complex system, especially for higher resolution system needs larger computation and cost for data. Meta-model technology which is based on being fitted through the simulation data, can used these data efficiently and support both the analysis of complex systems and the application for high-level simulations. It has become a emerging hotpot in the field of system simulation. Using meta-model, the high-level decision-makers can not only achieve the data support in good time, but also understand the system behaves more easily. In inter-hierarchy multi-resolution simulation, the technology that the meta-model, as a substitution for the low-order simulation model, is embed into high-order model can not only provide the data support for high-level model efficiently but also simplify the system so that the experiment efficiency is improved and it is possible to reuse the low-level higher resolution models in terms of data.The fitting approach of meta-model is the foundation of the meta-model application. However, how to choose the type and form of meta-model is a difficult job. This thesis discusses the fitting method for meta-model, especially the engineering generating approaches of meta-model, new fitting approach, applied strategy for choosing the type and form of meta-model, application of meta-model, etc.The main content and innovation of this thesis includes:1,The thesis expatiates the generating approach of meta-model, establishes an uniform frame of fitting approaches and the engineering methods of meta-modeling. This thesis concretely presents a generating method of meta-model using support vector machine.2,From application perspective, this thesis presents two strategies for solving the problem for choosing the type and form of meta-model. They are the approach of comparison of different meta-modeling methods, as well as the hybrid meta-modeling approach. This thesis concretely presents two new kinds of hybrid meta-modeling approaches.This thesis presents the hybrid meta-model method with error compensation in order to solve the equilibrium problem between the fitting precision of the metamodel and the conciseness and transparency of model; presents different fitting approaches convex linear combination prediction model in order to improve the precision of the metamodel and the flexibility of model. Through the theoretical analysis and experimental validation, the hybrid metamodel has better flexibility and application effect than a single metamodel. Meanwhile, a design method is presented for comparing the metamodel fitting approaches. Comparative analysis and application validation are made of several fitting methods such as polynomial regression, radial basis function, Kriging method, support vector machine, and so on.3,A support vector machine metamodel approach with state-embedded is presented for the system simulation mixed by continuous/discrete response.This thesis naturalizes the metamodelling problem of system simulation mixed by continuous/discrete response into a special high-dimension data fitting problem and builds the corresponding support vector machine metamodel by considering the discrete event as state variable. The experiment indicates that this metamodel has several merits such as high structure fidelity, high fitting precision and convenience for application.4,By means of the comparison of fitting approaches of meta-model and the hybrid meta-model method, a meta-model is constructed for detecting and tracking target in radar system of missile attack-defense simulation.The structural and computational complexity of radar system's simulation model is a main component of the complexity of missile attack-defense simulation experiment. We have constructed several types of meta-models for it. Through the experimental comparison, we have obtained a second-order polynomial meta-model, which is characterized by high fitting precision, rapid calculation, model conciseness, etc. The meta-model is used instead of the radar system simulation model to improve the efficiency of the experiment greatly in the missile attack-defense simulation.5,With regard to the resource distribution optimization analysis applicable to the military and economy, this thesis suggested a solution using meta-model. We illustrate the method of meta-model for solving the complex system analysis and inter-hierarchy multi-resolution simulation problems from the perspective of practice.It is hard to solve the resource distribution optimization analysis problem by the traditional mathematical method. Take the problem of linear programming optimization analysis and more complicated linear fractional programming optimization analysis for example, we analyze the problem theoretically and compare the fitting methods of meta-model by different experimental designs, correspondingly obtain the approximated meta-model of their optimal value function with higher precision and less calculation time. This method is applicable to the problem of non-linear programming optimization analysis. This thesis presents a linear programming optimization analysis with restriction condition, and analyzes the problem theoretically. Furthermore, it presents a solution using meta-model. This thesis presents the resource distribution multilevel optimization analysis problem. The technology that the meta-model is used to fit the low-order simulation model and Embed into high-order model, can reduce the cost greatly, and enhance the validity of decision making.