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 Dempster-Shafer evidence theory is used to detect and diagnose the calibration network in a control system by combining multiple neural network classifiers. A single neural network classifier classifies certain specific feature quantities. The generalized distance between the network output value corresponding to the actual system feature quantity and the network output mean value of the corresponding training feature set is that the actual system output by a single classifier belongs to a certain class Of the measure. The evidence theory adopts the form of evidence combination under the assumption of simple support set, and the final output is the result of synthesizing the output of multiple neural networks. The practical application shows that this method can detect and diagnose faults which can not be found by a single classifier and reduce the inaccuracy when using a single classifier to detect and diagnose different faults