基于随机LPNN网络的优化设计研究

拉格朗日乘子神经网络是一种适合于求解一般约束问题的神经网络。网络运行中附加动量项和引入逐渐衰减的高斯噪声。附加动量项方法能减少震荡时间,提高网络的收敛速度。高斯噪声能避免神经网络收敛于假吸引子,改善全局寻优能力。用该方法解决飞机总体参数优化问题。数值结果表明,算法的稳定性、全局寻优性、约束的满足程度好,同时拉氏乘子可以帮助进行最优设计结果的灵敏度分析。

Optimal Design Based on Stochastic Lagrange Programming Neural Network

The Lagrange programming neural network(LPNN) is good at solving general constraint problems.For the purpose of operating the neural network,momentum items are added and the attenuating Gaussian noise is introduced.The momentum item addition method reduces vibration time and accelerates the convergence of the neural network.The Gaussian noise introduction method prevents the neural network from converging to a pseudo-attractor and improves the ability to seek global optimization.These methods are applied to optimizing the overall parameters of an aircraft.The numerical results show that the algorithm for the methods has good stability,global optimization and constraint satisfaction.At the same time the Lagrange multiplier helps perform the sensitivity analysis of optimal designs.