基于GA-BP贝叶斯算法的可靠性分析近似模型

Approximation model for reliability analysis based on GA-BP Bayesian algorithm

研究了GA-BP(Genetic Algorithm-Backpropagation)贝叶斯算法在可靠性仿真中的应用.GA-BP贝叶斯算法是一种新型前馈神经网络训练算法,它建立在遗传算法(GA)、L-M(Levenberg-Marquardt)BP算法以及贝叶斯方法这三者的基础上.由于该算法的训练目标是获取对应于后验分布最大值的权值向量,并且在搜索过程中融入了遗传算法,因此能够使前馈神经网络具有更佳、更稳定的泛化性能.在可靠性仿真中,采用GA-BP贝叶斯算法来构造前馈神经网络近似模型,再用它来替代复杂费时的数值仿真程序进行Monte Carlo模拟,就能够在计算成本得到有效控制的同时获取随机输出变量的概率分布情况.

The usefulness of genetic algorithm-backpropagation（GA-BP） Bayesian algorithm was studied and evaluated for reliability simulation. GA-BP Bayesian algorithm is a algorithm to train feedforward neural networks, and it is based on GA, L-M （Levenberg-Marquardt） BP, and Bayesian method. The algorithm trains a network with the purpose of obtaining the weights corresponding with maximum posterior probability, and it adopts genetic algorithm in searching process. As a result, it makes neural networks have better and steadier generalization ability. When running a reliability simulation, GA-BP Bayesian algorithm can be uti- lized to train neural networks to make an approximation model that can be used in Monte Carlo simulation in- stead of expensive numerical program. In this way, the probability distribution of random ouput variables can be obtained with efficiently-controlled computing cost.