基于过滤器-封装器组合模型的故障特征选择算法

A Novel Hybrid Filter-Wrapper Algorithm of Feature Selection for Fault Diagnosis

利用过滤器法运行速度快和封装器法精度高的优点,提出了一种新的Filter-Wrapper两阶段组合式特征选择算法。在Filter阶段,算法采用Fisher准则对特征进行排序;在Wrapper阶段,以分类器的性能作为适应度函数,根据特征排序结果采用遗传算法搜索特征子集。运用滚动轴承故障模拟试验的数据对所提出的算法进行了验证,结果表明,相比单一的过滤器法或封装器法,该算法能够同时提高特征选择的性能和效率。

A novel hybrid feature selection algorithm of filter and wrapper was proposed,that took advantage of both approaches. It began by the feature ranking procedure based on Fisher＇s criterion （filter approach） to instruct the initialization of genetic population, then it run genetic algorithms （wrapper approach） to search optimal feature subset with the performance of classifier for evaluation function. The proposed method was applied to fault data sets of rolling bearing fault simulation experiment. The result shows that it can get high accuracy while maintaining high efficiency compared with each individual feature selection method.