参数优化的支持向量机机车车轮状态检测

Parameter optimization of support vector machine for locomotive wheel state detection

针对车轮状态检测中存在的类间样本误分代价不等的问题,利用自适应变异粒子群算法对代价敏感支持向量机(CS-SVM)的参数进行优化。该方法从数据样本中提取不均衡数据创建训练样本,采用代价敏感的支持向量机建模。为了提高分类的精确度,选用径向基核函数优化模型结构。提出了自适应变异粒子群算法优化CS-SVM的两个不同惩罚参数和核函数,并用参数优化的代价敏感支持向量机实现重载机车车轮状态分类。最后,通过仿真验证,车轮状态检测的平均准确率可以达到95%,平均处理速度24 s,具有实时性和较强的鲁棒性,能够满足重载机车运行要求。

An adaptive mutation particle swarm optimization algorithm for cost sensitive support vector machine（ CS-SVM） is presented for the wheel state detection in the presence of inter class sample misclassification cost problem of unequal. The method from sample data extracted from imbalanced data to create the training samples,using cost sensitive support vector machine modeling. In order to improve the accuracy of the classification,choose radial basis kernel function to optimize the structure of the model. The adaptive mutation particle swarm optimization algorithm is proposed to optimize the two different penalty parameters and kernel function of CS-SVM,and the cost sensitive support vector machine is used to realize the state classification of heavy load locomotive wheels. Finally,the simulation results show that the average accuracy of wheel state detection can reach more than95%,the average processing speed of 24 seconds,with real-time and strong robustness,can meet the requirements of heavy load locomotive.