基于CRPF的残差似然比检验故障诊断算法

Residual likelihood ratio test for fault diagnosis based on cost reference particle filter

分析了非线性系统故障诊断领域中常用方法的优缺点,针对外界随机扰动对于滤波精度的不利影响以及故障诊断的连续实现问题,通过代价评估的粒子滤波、交互式多模型和序贯概率比检验三者的有机结合,提出了一种基于代价评估粒子滤波的残差似然比检验故障诊断算法。采用代价评估粒子滤波替代交互式多模型中的次优滤波器,同时简化交互式多模型输入交互和输出交互环节。将滤波过程中得到的残差信息引入序贯概率比检验框架中,构建了一种新的在线残差似然比检验方法。实现了对于非线性系统状态有效估计以及对于系统模式连续、可靠的辨识。计算机仿真实验验证了算法的有效性。

The existing problems of common methods in fault diagnosis are briefly analyzed. In view of the adverse effect of external disturbance and the requirement of the successive implementation, by the triple integration of the cost reference particle filter, the interacting multiple model and the sequential probability ratio test, a novel residual likelihood ratio test algorithm based on cost reference particle filter for fault diagnosis is proposed. First, the cost reference particle filter is used to substitute the suboptimal filter in interacting multiple models, and the input interaction step and the output step are simplified. Then, the residual information is introduced into the sequential probability ratio test frame to construct an online residual likelihood ratio test method. The new algorithm realizes the efficient estimation for system state and the successive and reliable identification for system models. Comouter simulation verifies the validitv of this algorithm