遗传算法和神经网络组合求解离散变量结构优化问题

How to determine discrete variables in structural optimization by neutral networks and genetic algorithm

利用遗传算法优良的全局搜索性能和对目标函数的仅要求有定义的特点进行离散变量结构优化设计。利用BP神经网络的模拟逼近功能,建立结构设计输入与输出之间的全局映射关系,获得遗传算法中的适应度函数值。将此方法应用于垂直循环式停车设备钢结构骨架优化问题。结果表明,该方法只需较少的有限元分析次数就可以获得良好的优化解。

The paper expounds how to perform the discrete structure optimization by making full use of the characteristics ot the excellent global search capability and only need in definition for the objective function from genetic algorithm, and how to oblain the fitness values using simulation function of BP networks and constructing a non - linear mapping function between input and output data. The application of this method to structural optimization in vertical circulating parking system shows that this method can obtain the better optimal result with less times of FEA.