基于改进PSO-LSSVM的脉冲涡流缺陷二维轮廓重构  

2D Profiled Reconstruction for PEC Defect Based on Improved Particle Swarm Optimization and Least Squares Support Vector Machine

作　　者：钱苏敏[1] 左宪章[1] 张云[1] 常东[1] 

机构地区：[1]军械工程学院无人机工程系,河北石家庄050003

出　　处：《仪表技术与传感器》2013年第8期99-102,共4页Instrument Technique and Sensor

摘　　要：基于最小二乘支持向量机(LSSVM)对脉冲涡流信号构建了缺陷二维轮廓重构模型。以脉冲涡流信号的垂直分量By的正峰值包络作为LSSVM重构模型的输入,缺陷的二维轮廓作为输出,对LSSVM进行网络训练。同时,利用改进的粒子群算法优化模型参数,提出模型泛化能力的评估方法。通过比较基本粒子群算法及改进的粒子群算法的LSSVM模型的重构结果,验证了改进粒子群算法的LSSVM重构模型的有效性。The 2D profiled reconstruction model was constructed for pulsed eddy current flaw signals with least squares support vector machine（LSSVM）.Take the positive peak envelope of the component By as the input of LSSVM with the method of magnetic flux density with PEC signals and 2D profile of defects as output and train for LSSVM network.The paper used improved particle swarm optimization to optimize the model parameters,and presented the assessment methods to model generalization capability.Comparing the reconstructed results of LSSVM model which uses either particle optimization swarm algorithm or improved particle swarm optimization algorithm,it improves the effectiveness of the LSSVM reconstruction model which uses improved particle swarm optimization algorithm.

关 键 词：二维轮廓重构 粒子群优化 混沌序列 最小二乘支持向量机 

分 类 号：TG115[金属学及工艺—物理冶金]