一种基于人工神经网络和优化算法库的复杂虚拟样机优化通用方法

A General Method Used in Complex Virtual Prototype Optimization Based on ANN and Optimization Algorithms Library

复杂虚拟样机仿真需要求解数量巨大的方程,单次运行就需要很长的时间,而优化时需要成千上万次地进行样机仿真,同时多个优化变量和多局部极值的优化目标函数也对优化算法提出了很高的要求,这就给复杂虚拟样机优化带来很大困难。分析了复杂虚拟样机优化的特点,提出了用神经网络拟合样机函数和用优化算法库选择优化算法的思想,形成了一种适用于复杂虚拟样机优化的通用方法。某型装备整体动态性能的优化实例证明了该方法的有效性。

There are many difficulties in complex virtual prototype optimization. First, optimization needs thousands of simulation which must solve numerous equations and use a very long time in just one cycle. In addition, many optimization variables and many local minimums are captious to optimization algorithm. After analyzing the characteristics of complex virtual prototype optimization, thoughts of fitting virtual prototype function with ANN and choosing optimization algorithm in optimization algorithms library are brought out, and a general method suitable to complex virtual prototype optimization is obtained. In the end, an example of certain equipment whole dynamic performance optimization proved the efficiency of the method.