LS-SVM辅助的小幅值及缓变故障检测与容错方法

LS-SVM assisted fault detection and tolerance method for small-amplitude fault and gradual fault

在INS/GNSS组合导航中,针对传统残差χ^2检验法对小幅值突变故障和缓变故障检测效率不高的问题,提出了一种最小二乘支持向量机(LS-SVM)辅助的故障检测与容错方法。该方法通过构建双阈值检测门限来判断观测数据质量,当观测数据受污染时,利用LS-SVM 对残差的预测值代替卡尔曼滤波残差自适应调节滤波增益矩阵,降低漏警故障对状态估计的污染,以提升估计精度和故障检测灵敏性。仿真结果表明,对于小幅值突变故障和缓变故障,相比于传统残差χ^2检验法,所提方法检测漏警率分别能降低25%和15%以上,故障期间滤波精度可提高85%以上。

Aiming at the problem that the low efficiency of traditional residual chi-square test in detecting small-amplitude fault and gradual fault of INS/GNSS integrated navigation, a fault detection and tolerance method assisted by LS-SVM is proposed. This method adopts double-threshold test to judge the quality of measurement. When the measurement is polluted by fault, the residual of Kalman filter is replaced by the forecasted residual of the least squares support vector machine regression to adjust the gain matrix adaptively, which can reduce the pollution of missed detected fault on the state estimation, and this in turn improves the estimating accuracy and the sensibility of fault detection. Simulation results show that, compared with the conventional residual chi-square test method, the proposed method can reduce the missed detection rates of small-amplitude fault and gradual fault by at least 25% and 16% respectively, and the filtering precision during the fault period can be improved by more than 85%.