基于近邻保持嵌入的卫星姿态控制系统微小故障检测

摘要： 以卫星姿态控制系统(ACS)为研究对象,以微小故障检测为研究目标,提出一种改进的近邻保持嵌入算法(EWMA DNPE).针对近邻保持嵌入(NPE)算法中邻域参数无法自动设定的缺陷,通过引入动态邻域,使得动态近邻保持嵌入算法（DNPE）可根据流形的样本密度动态地选取近邻点计算权值重构矩阵.将指数加权移动平均(EWMA)引入DNPE,通过EWMA对历史故障数据的累加作用,建立SPE统计量实现微小故障的检测,并仿真验证了EWMA DNPE算法对卫星ACS微小故障检测的有效性和可行性.

Incipient Fault Detection Method of Satellite Attitude Control System Based on Neighborhood Preserving Embedding

Abstract：In this paper, an advanced neighborhood preserving embedding method combined with exponentially weighted moving average (EWMA DNPE) is proposed in incipient fault detection for satellite’s attitude control system (ACS).Considering the artificial set of neighborhood parameters in neighborhood preserving embedding (NPE) algorithm, the dynamic neighborhood selection is introduced, so that the weights reconstruction matrix can be selected according to the sample density of the manifold. On the basis of this, the EWMA DNPE algorithm is used to establish the squared prediction error (SPE) statistic in fault detection (FD) by exponentially weighted moving average (EWMA). The case study on a simulated satellite ACS demonstrates the effectiveness and feasibility of the proposed method.