Neural Networks Based Methodology Research On Simulation Metamodels

In order to reduce the complexity of experiments in anti-ship missile system simulation, this paper proposes the idea to build meta-model in form of neural network for subsystem and to embed the meta-models in the experiment framework for system simulation. The theories and characteristics of three kinds of neural network are analyzed for meta-modeling. Some critical issues on selecting network structure and training network for The BP (Back-Propagation) neural network experiment design method in meta-modeling are discussed. Finally, a meta-model for air surveillance radar in anti-ship missile weapon system simulation, is created. By analyzing the simulation data generated by meta-model, the method is validated to be effective. The main jobs of this thesis are as follows:(l)The meta-model framework research, which focuses on meta-models modeling^ evaluating and applying, is made;(2) The idea of modeling meta-models in complex weapon engagement simulation with BP network is brought forward;(3)Some critical issues on building meta-models and validating the corresponding solutions are elaborated;(4)The application software for meta-modeling, which can improve the function of the experiment analyzer component in Sim2000, is implemented.