中文摘要：

遗传算法直接应用于结构优化设计时，需要对每代进化种群的全部个体进行结构有限元分析。这将耗费大量的计算时长．文章针对这一问题，首先采集若干组进行结构有限元分析的数据样本，并代入BP神经网络进行网络训练和仿真，然后在运用优化算法进行优化迭代运算时，回避大量个体的有限元重分析过程，而代之以BP神经网络的预测结果，从而大量节省了计算时间．17杆平面桁架结构和42杆空间结构的算例表明，采用本文方法的计算耗时分别比原始的遗传算法节省80．39％和83．21％，且优化迭代过程能够稳定收敛，从而验证了本文方法的有效性．

英文摘要：

The genetic algorithm is a commonly used method for structural optimization. This method requires to re-analyze the structure acted by loads for every individual in every evolution. Obviously this process is at a huge cost of computation. In this study, the BP neural network is employed to modify this procedure. That is, some special training samples of structure are analyzed by the FEM software, and then, the results of nodal displacement are used to train the BP network. After that, in the evolution of a GA, re-analyzing of the structure is no longer required, and the structural displacements are predicted by the trained BP network. This procedure requires much less computing time than the original GA. Two examples, including a 2D truss structure with 17 members and a 3D truss structure with 42 members are presented to show the validity of the proposed method. 80. 39% and 83.21% of computing time are saved by applying the proposed method, which demonstrates the validity of the method.