ReliefF与QPSO结合的故障特征选择算法

The fault feature selection algorithm of combination of ReliefF and QPSO

为提高故障数据集的分类精度,将ReliefF算法与量子粒子群算法(Quantum Particle Swarm Optimization,QPSO)进行结合,提出一种能够降低故障数据集维度的敏感故障特征选择方法。首先,在对经滤波消噪后的故障信号进行多域量化特征提取基础上,设定时域与频域特征、经小波包分解得到的各频带能量特征作为描述转子系统故障状态的初始故障特征集,并用转子系统的典型故障模拟信号集合得到了一种原始的故障数据集。随后,用ReliefF算法通过迭代计算得到的权值对故障数据集各特征向量进行加权、并设定阈值剔除不相关特征,据此实现了对原始故障数据集各特征的第一次筛选。最后,引入量子粒子群算法(QPSO)对特征集合进行二次筛选,剔除不利于实施分类的冗余特征并同时实现优化支持向量机的参数,通过处理得到了一种精简的最优特征子集和最合适的一组支持向量机参数。用得到的原始故障数据集对所建立的方法性能进行了计算验证。结果表明,该方法可有效地筛选出规模较小且故障模式辨识度高的低维故障数据集,它可显著提高故障分类器的辨识准确率。

In order to improve the classification accuracy of fault data sets,the combination of ReliefF algorithm and Quantum Particle Swarm Optimization(QPSO)was adopted to propose a sensitive fault feature selection method that can reduce the dimension of fault data sets.First of all,on the basis of multi-domain quantitative feature extraction of the filtered and denoised fault signal,setting the time domain and frequency domain features,the energy features of each frequency band obtained by wavelet packet decomposition were set as the initial fault feature set to describe the fault state of the rotor system.The typical fault of rotor system analog signal collection got a primitive failure data set.Then,the weight obtained through iterative calculation by the ReliefF algorithm was used to weight each feature vector of the fault data set,and the threshold was set to eliminate the irrelevant features.So as to realize the first screening of each feature of the original fault data set.Finally,a Quantum Particle Swarm Optimization(QPSO)algorithm was introduced to filter feature sets twice,eliminating redundant features that were not conducive to classification and optimizing the parameters of support vector machines at the same time.So a simplified optimal feature subset and the most appropriate set of support vector machines parameters were obtained.The method performance was verified by the original fault data set.The results show that this method can effectively screen out low-dimensional fault data sets with small size and high fault pattern recognition,which can significantly improve the identification accuracy of fault classifier.