基于Fuzzy ART的航天器状态监测方法及应用研究

Research on Application of Fuzzy ART Neural Network to Condition Monitoring for Spacecrafts

为解决信号空间法在航天器状态监测中存在的参数阈值难以设定问题,提出一种基于FuzzyART神经网络的状态监测方法.该方法利用FuzzyART的聚类特性及网络的自主扩展性,对系统正常状态向量在多维空间进行聚类,并以网络的输出节点编号代表正常状态向量的类别.监测时,网络通过判定输入的状态向量与正常状态向量的匹配度是否满足要求而给出监测结果.该方法克服了信号空间方法各单项参数阈值难以设定问题,并且能有效地消除因噪声干扰产生的虚警.

Condition monitoring method based on Fuzzy ART neural network is presented for solving the difficulty of setting threshold value of parameter while signal-space method is used in condition monitoring of spacecrafts. The normal state vectors are clustered in multiple dimension space utilizing Fuzzy ART clustering characteristic, the autonomous expansibility of network output-layer node, and the index number of network output nodes, which represents the classification of normal condition vector. While monitoring, the network obtains the monitoring result through judging that whether or not the matching degree between input state vector and normal state vectors meets challenge. The method overcomes the problem that the threshold value of each parameter is hard to confirm for signal space method, and it can reduce false alarm because of noise interference. The method is validated in a real test.