摘要
讨论了利用多粒子群优化算法(Multi-PSO)和径向基函数(RBF)神经网络进行缺陷参数红外识别的途径。PSO算法可以不用计算梯度,算法通用,而使用RBF神经网络作为代理模型,极大简化了复杂、费时的有限元计算,其中训练RBF神经网络的样本由有限元软件的计算结果产生。提出的多粒子群优化算法将粒子群分为若干子群,并利用粒子本身、粒子所在子群以及全局的最优解来更新粒子的速度与位置,该方法收敛速度较慢,但有可能找到问题的多个极小值。最后给出了该方法在缺陷参数红外识别中一个简单的应用例子。
A multi-particle swarm optimization （multi-PSO） algorithm, combined with an RBF neural network is applied in the defect identification of infrared non-destructive test/evaluation （NDT/E）. The PSO algorithm is derivative free. And the surrogate model of radial basis function （RBF） neural network with the samples generated by the finite element method （FEM） solver, is a precise substitution of the complex and time-consuming FEM computation. The proposed multi-PSO algorithm divides a swarm into several sub-swarms. And the velocity and the position of each particle are updated by the optimums of itself, the sub-swarm and the global. It makes the convergence more slowly but more optimum of the problem can be found. Finally, a simple validation case of defect identification in infrared NDT/E is provided.
出处
《数据采集与处理》
CSCD
北大核心
2008年第B09期66-72,共7页
Journal of Data Acquisition and Processing
基金
总装“十一五”装备维修改革基金(KY38010914)资助项目
关键词
粒子群
优化算法
缺陷识别
径向基函数
神经网络
particle swarm
optimization algorithm
defect identification
radial basis function
neural network
