基于健康度分析的卫星故障预测模型  

Satellite fault prediction model based on health analysis

作　　者：朱昶文 党建成[1] 周军[1] ZHU Changwen;DANG Jiancheng;ZHOU Jun(Shanghai Institute of Satellite Engineering,Shanghai 201109,China)

机构地区：[1]上海卫星工程研究所

出　　处：《计算机应用》2019年第S02期34-40,共7页journal of Computer Applications

基　　金：中国航天科技集团公司第八研究院第五〇九研究所卫星在轨管理保障条件建设项目(ZKJ0080509-16-B12048)

摘　　要：针对目前卫星监测系统无法及时发现故障且无法反映卫星状态趋势变化的问题,提出一种基于自适应白噪声完整集成经验模态分解(CEEMDAN)方法与粒子群极限学习机组合(PSO-ELM)的指标预测模型和基于模糊层次分析法(FAHP)的多指标融合故障检测模型。首先,通过CEEMDAN算法对各指标进行分解后通过PSO-ELM预测;其次,对各指标预测值分别建立非线性无量纲模型,得到各指标“健康度”;最后,利用FAHP法对融合各指标“健康度”得到组件“健康度”,利用“健康度”判断是否发生故障。通过某在轨卫星蓄电池组故障数据实验可得,CEEMDAN-PSO-ELM预测模型在平均绝对误差(MAE)、均方根误差(RMSE)指标上分别为0.1029和0.125,均优于文中提到的其他预测模型;该故障检测模型与目前卫星监测系统相比,能提前两周期检测到故障。最后一句意义不大,作者同意删实例验证表明,该模型能实现故障预测功能。An index prediction model based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)and Particle Swarm Optimization(PSO)-Extreme Learning Machine(ELM)and a multi-index fusion fault detection model based on Fuzzy Analytic Hierarchy Process(FAHP)were proposed for the problem that the current satellite monitoring system cannot find faults in time and cannot reflect the trend change.Firstly,the CEEMDAN algorithm was used to decompose the indicators and predicted them separately by PSO-ELM algorithm.Secondly,a nonlinear dimensionless model was established for each indicator s predicted value,and the health of each index was calculated.Finally,the FAHP algorithm was used to integrate the health of each index to form component health,and compared by the preset threshold to determine whether a fault has occurred.For on-orbit satellite battery fault data,the CEEMDAN-PSO-ELM prediction model achieved 0.1029 and 0.125 in MAE and RMSE indicators,better than the other prediction models mentioned.Compared with the current satellite monitoring system,the proposed fault detection model can detect faults two cycle ahead of time.

关 键 词：故障预测 自适应噪声完整集成经验模态分解 粒子群优化算法 极限学习机 模糊层次分析法 

分 类 号：TP206.3[自动化与计算机技术—检测技术与自动化装置]