论文部分内容阅读
为了利用不断积累的网络样本提高故障诊断效能,针对标准支持向量机不直接支持增量学习的问题,提出一种边界偏转覆盖增量支持向量机.根据违背Karush-Kuhn-Tucker条件的新增样本在特征空间中可引起原分类边界改变的情况,设计边界偏转覆盖算法预选支持向量再生区作为增量训练工作集,解决了难以确定的非支持向量向支持向量的转化问题.理论分析和实验结果表明,该方法能有效简化训练工作集,在保证故障诊断精度的同时大幅度提高增量训练效率.
In order to improve the performance of fault diagnosis by using the continuously accumulated network samples, aiming at the problem that standard support vector machines do not support incremental learning directly, a new incremental support vector machine with boundary-deflection covering is proposed. According to the new sample which violates the Karush-Kuhn-Tucker condition The original classification boundary can be changed in the feature space, and the boundary-rotation coverage algorithm is pre-selected as the training set of incremental training to solve the difficult problem of transformation from non-support vector to support vector.Theoretical analysis and experimental results It shows that this method can effectively simplify the training set and greatly improve the efficiency of incremental training while ensuring the accuracy of fault diagnosis.