基于异构粗糙神经网络集成的故障诊断研究

将粗糙集与神经网络集成相结合，提出一种基于异构粗糙神经网络集成进行故障检测的方法。首先利用粗糙集的属性约简能力，从给定数据集中去除冗余信息；然后基于负相关学习理论构造多个异构成员神经网络，最后组合多个训练好的异构神经网络进行故障诊断。该方法不仅显著提高了神经网络的泛化能力，而且无需预先确定神经网络的拓扑结构，简单易用。设计了四种不同的诊断器在柴油机供油系统的标准样本集上进行的诊断测试实验，结果表明，基于异构粗糙神经网络集成的故障诊断方法具有最好的诊断正确率。

Fault Diagnosis Based on Heterogeneous Rough Neural Network Ensemble

A new fault diagnosis method called heterogeneous rough neural network ensemble (HRNNE) is presented in this paper, which combines the rough set and the neural network ensemble to provide outstanding ability to diagnose faults. In HRNNE, rough sets theory is used to simplify attribute parameter and eliminate unnecessary property, and a heterogeneous neural network which consists of several component neural networks is built with negation correlation to determine the faults. HRNNE can improve the generation capacity of the neural networks evidently, but also need not to determine the structures of the neural networks beforehand. It is very easy to use in practice. We design four different diagnosis classifiers to compare the diagnosis performance in the standard sample set of the fuel injection system of diesel engine. The experiment results show that HRNNE is the best one, which get high diagnosis accuracy.