Anomaly Detection Method of Space Payload Using Multivariate State Estimation Technique and Self-Organizing Feature Map

Abstract

Anomaly detection technique is important for finding abnormal parameters and potential faults of equipment. Under the influences of operating instructions, environment conditions and equipment performance, the parameters of space payload fluctuate obviously and traditional threshold value method is unable to ensure the accuracy of anomaly detection. The paper proposed a novel anomaly detection method based on unsupervised learning and time series analysis, thus a new perspective of data correlation and data evolution method are achieved. The proposed method includes offline training process and online detecting process. The offline training process utilizes amounts of normal sampling to establish the normal state model based on multivariate state estimation technical (MSET). To guarantee the accuracy of normal state model, self-organizing feature map (SOM) is introduced to optimize modeling accuracy of memory matrix establishment for MSET, thus the normal model established in offline training process could cover overall working conditions. In online detecting process, the established normal model based on MSET could predict the objective parameter and the deviation of actual data and predict data could reflect the abnormal state of equipment. Finally, the effectiveness of the method is verified by experiments. The experiments result shows that the proposed method could detect the abnormal state of space payload accurately and rapidly.