论文部分内容阅读
提出了一种新的学习策略,用于解决发动机故障诊断中大规模支持向量机(SVM)的训练问题.通过保留初始SVM分类器支持向量超平面附近的样本以及错分样本,使最终得到的约减集规模明显缩小,从而可在保持较高分类精度的前提下使训练时间明显缩短;同时,由于支持向量的数量减小,分类时间也相应缩短.探讨了序贯最小优化(SMO)算法的参数选择和实现过程中的关键问题,为这种极具潜力的算法在发动机故障诊断中的实际应用奠定了坚实的基础.仿真实例表明,这种基于大规模训练集SVM的发动机故障诊断方法有效、可靠,容易实现,可以作为工程应用的基础.
A new learning strategy is proposed to solve the training problem of large scale support vector machine (SVM) in engine fault diagnosis.By keeping the samples near the hyperplane of the initial SVM classifier and the misclassified samples, Reduce the size of the approximate reduction set significantly reduce the training time can be significantly reduced under the premise of maintaining a high classification accuracy; the same time, due to the reduced number of support vectors, the classification time is also reduced corresponding sequential minimum optimization (SMO) algorithm The key problems in parameter selection and implementation of this method lay a solid foundation for the practical application of this highly potential algorithm in engine fault diagnosis.The simulation examples show that this kind of SVM-based engine fault diagnosis method Effective, reliable, easy to implement, can be used as a basis for engineering applications.