组合多重神经网络动态系统鲁棒故障检测与诊断

利用Ｄｅｍｐｓｔｅｒ－Ｓｈａｆｅｒ证据理论，通过组合多重神经网络分类器，对一控制系统中的校正网络进行故障检测与诊断．单个神经网络分类器对某些特定的特征量进行分类，对应实际系统特征量的网络输出值与相应训练用特征集的网络输出均值之间的广义距离为单个分类器输出的实际系统属于某类的度量值．证据理论采用简单支撑集假设下的证据组合形式，最终的输出为综合多个神经网络输出后的结果．实际应用表明，此方法可以检测与诊断出单一分类器不能发现的故障，同时也减少了利用单个分类器对不同故障进行检测与诊断时的不精确性

Detection and Diagnosis of Robust Faults of Dynamic System by Combining Multiple Nerve Networks

The fault of rectifying network were detected and diagnosed by using the evidence theory and combining multiple nerve network classifiers,Single nerve classifier was used to classify the specified features.The general distance between the network output corresponding to real systems feature and the expected value corresponding to the feature set used for the training of the classifier is a measurement which belongs to a real systems output.The evidence theory was used in a combination form based on the assumption of simple support set concept,and the final conclusion is the combined output of multiple nerve network classifers.Practical applications have proved that the method can detect and diagnose faults that single classifier cant find and can reduce inaccuracy which occurs when single classifier is used to detect and diagnose different faults.