中文摘要：

利用自学习粒子群优化算法的全局寻优能力克服梯度下降法过分依赖初始解,易陷入局部极值的缺点,从而提高梯度下降法的优化性能。以径向基函数神经网络为前向模型,基于自学习粒子群与梯度下降混杂的反演方法用于漏磁缺陷轮廓重构中。实验结果表明,该反演方法重构的缺陷轮廓比较准确,且在漏磁信号存在噪声的情况下,重构结果到与实际轮廓相近,并具有一定的噪声鲁棒性。

英文摘要：

By using the global optimization ability of self-learning PSO,overcome the disadvantages of gradient descent method unduly relying on initial solution and easy falling into local minimum,thereby the optimizing performance is enhanced. Radial basis function neural network is used as forward model,and the inversing approach based on the hybrid of self-learning PSO and gradient descent is applied to reconstruction of defect profile on magnetic flux leakage. The experiment results show the profiles reconstructed by the proposed approach are relatively accurate and still close to the true profile with existence of noise in magnetic flux leakage signal. So the proposed approach is partly robust to noise.