LS-SVM method for 2-D Reconstruction of the Oil Pipeline Defect Based on PSO algorithm

Abstract

2-D (two-dimensional) reconstructing of the oil pipeline defect from MFL (magnetic flux leakage) signals is a difficult problem in MFL testing. The traditional solution is based on the neural network (NN) algorithm. But it has the disadvantages of complex structure, slow speed and low accuracy. To improve these disadvantages, a LS-SVM (least squares support vector machine) method is proposed for reconstructing the defect based on PSO (particle swarm optimization) algorithm in this paper. LS-SVM algorithm instead of traditional NN algorithm is used to overcome the problems such as local minimum point, curse of dimensionality and over-fitting. The calculation is simplified meanwhile. PSO algorithm is used to optimize the regularization parameter and kernel parameter of LS-SVM, and improve the accuracy of reconstruction. The simulation and experimental results show that, compared with the traditional reconstruction methods, this new method can indeed get better reconstruction effect with higher accuracy and faster processing speed.