摘要： 试用离线训练的神经网络进行导航传感器故障检测。首先用从某船试航时的一段数据中选出的包含多种航行状态的数千组平台罗经读数训练神经网络并同时选择神经网络的输入延迟数和隐层单元数。然后用已选择好结构并训练好的神经网络作为在线估计器对平台罗经的读数进行一步预测。最后根据平台罗经的读数与其对应的预测值之间的差值进行故障检测。分别用同一艘船同一次试航的两段平台罗经读数进行训练和故障检测仿真,结果证明该方法可行。

关键词: 神经网络, 故障检测, 在线估计器, 离线训练, 导航, 传感器

Abstract: In this article, neural networks are tried to perform navigation sensor failure detection. In the proposed approach, first, neural networks are trained and the numbers of input delay and hidden units are selected simultaneously with off-line sample in which there are several thousands sets of platform compass readings which cover several navigation status. And then, the selected and trained neural networks are used as on-line estimators to predict the next sensor readings. The discrepancy between the sensor readings and their estimation is used to detect sensor failures. Simulation is performed with platform compass readings of another section of the voyage. The simulation result shows the proposed approach is feasible. ;;;;;

中图分类号: 

• TP183

• TP212