电控发动机故障诊断属性约简算法应用研究

利用粗糙集理论中的区分矩阵属性约简算法对电控发动机的几种典型故障参数进行属性约简,为验证约简结果是否有利于下一步的故障诊断,采用较成熟的BP神经网络对其进行诊断验证,将约简结果作为网络的输入,待诊断故障作为网络的输出。通过学习训练结果表明:利用区分矩阵方法所获得的核约简不能作为故障诊断的特征参量,其导致网络不收敛,而其它3组约简可以用于区分现有故障。为获得最优的属性约简结果,利用二进制粒矩阵的方法进行了最优属性约简的证明。

Application Research on Attribute Reduction Algorithm for Fault Diagnosis of Electronic-controlled Engine

Several typical faults parameters of electronic-controlled engine were taken as an example to do attribute reduction using discernible matrix of rough set theory. In order to validate reduction results whether it helps the next fault diagnosis, the relatively mature BP neural network was chosen to test and verify. The reduction results are as the inputs of BP neural network, and the typical faults are as outputs. The training results show that the received core reduction using the discernible matrix method cannot be used as the characteristic parameters of fault diagnosis, it does not cause network convergence, and the other three groups of reduction can be used to distinguish the existing faults. In order to get the best attribute reduction, the optimal attribute reduction was proved by using Bit Granular Matrix.