改进粒子群优化的多类LS-SVM电机故障识别算法

Fault Diagnosis Algorithm of Multi-LS-SVM Classifier Electric Machine Based on Modified PSO

传统电机故障诊断方法具有不确定性。多类LS-SVM方法所需样本较少、识别率高,可有效识别电机故障,但该方法计算过程中有庞大的矩阵求逆运算。为降低计算数据复杂度、提升训练速度,提出基于改进粒子群算法的电机故障识别算法。该算法依据种群收敛程度与个体自适应值调整惯性权重,选择一对余的LSSVM多分类器结构,构造4个改进粒子群的多类LS-SVM分类器,识别电机4类故障。实验验证表明,该算法可保证电机故障识别结果准确率,具有较好的实用性和推广性。

The traditional motor fault diagnosis method has uncertainty.The multi class LS-SVM method needs less samples and has high recognition rate,which can effectively identify motor faults.However,there is a huge matrix inversion operation in the calculation process of this method.Aiming at adjusting inertia weight and thus accelerating the training speed through convergence of swarm and in⁃dividual fitness,this article proposes the modified PSO.By iteratively solving the matrix in LS-SVM through adaptive PSO,the prob⁃lem of solving inverse matrix is avoided and the memory is saved.In order to distinguish between 4 faults for electric machine quickly and accurately,one-against-rest LS-SVM multiple classifier structure is chosen in the model to build 4 LS-SVM classifiers based modified PSO.Diagnosis test results show that the proposed method has high classification accuracy,which proves its effectiveness and usefulness.