基于Hopfield网络的飞机设计

Hopfield神经网络与增广拉格朗日乘子法相结合来求解非线性约束优化.神经网络作为求解乘子法的子问题的动力学方法,仅需计算一阶导数.引入逐渐衰减的高斯噪声信号构造随机神经网络.同时针对随机网络受初始温度制约,跳出局部极小值能力有限的问题,网络运行采用结合模拟退火的欧拉法.用该方法对某喷气教练机进行总体优化设计,结果表明,算法的数值稳定性较好,求解精度高.并基于拉氏乘子提供的约束敏度信息,做了设计要求权衡.最后研究了某型干线旅客机的机翼气动/结构综合设计问题.

Hopfield network based approach to aircraft design

Combining augmented lagrange multiplier(ALM) method to Hopfield neural network(HNN),was proposed to solve nonlinear constrained optimization.HNN is taken as a dynamic approach for minimization subproblem in ALM method,only needing the first derivative of the Lagrange function.The random neural network was extended by adding Gaussian noise gradually reducing with the temperature,whose ablity escapeing from the attraction of the localminimum points was limited by the initial temperature.Combined with the simulated annealing,an improved algorithm for Euler method was presented for numerical implementation of the network.The approach was applied to jet trainer aircraft preliminary optimization.The results show that the computing process is stable and the optimal result is enough precise.The trade-off of design requirements was studied through the Lagrange multipliers.An aerodynamic/structural design optimization for the wing of a trunkeliner was studied.