卫星姿控系统基于模糊基函数网络与自回归模型的故障预测

Fault prognosis in control system for satellite attitudes based on fuzzy basis function networks and autoregression model

针对卫星姿态控制系统的故障预测问题,给出了模糊基函数网络(FBFN)与自回归模型(AR)相结合的故障预测方法,并提出了预测置信因子的概念,对故障预测的准确性进行评价.首先利用卫星正常运行时的姿态数据训练FBFN,将训练好的FBFN作为卫星姿控系统的标准输出模型;然后把卫星实时姿态数据与FBFN输出数据之间的差值作为残差,利用AR模型对残差序列进行建模,进而对未来的残差进行预测;最后依据预测残差的统计分布给出了故障发生概率,利用故障预测置信因子来描述预测步长不同时故障预测结果的可信性.

A new method based on fuzzy basis function networks（FBFN） and autoregression（AR） model is proposed for predicting faults in the control system for satellite attitudes.Firstly,normal satellite attitude data are used to train FBFN which is used as the standard model of the control system for satellite attitudes.Secondly,the real-time attitude residual errors are obtained by subtracting the FBFN output from the real-time data of satellite attitudes.Thirdly,the time series of the residual errors is used to build an AR model.Therefore,the faults in the control system for satellite attitudes are predicted by using the AR model,and the failure probability is given according to the statistical distribution of the prediction errors of the AR model.Finally,the confidence factor is determined which shows the confidence measure of the fault prognosis.