噪声方差自适应修正的混合系统故障诊断方法

针对混合系统故障诊断问题，提出了一种模型噪声方差自适应修正的多模态故障诊断方法。首先，在粒子滤波的框架内将混合系统故障诊断建模为最优状态估计与跟踪问题，利用实时观察信息和各个模态先验的转移概率，估计最优的故障模态，并针对估计结果进行单独的建模分析；接着，根据平滑估计值和当前观测信息之间的相关性，建立噪声方差在线自适应检测机制，对模态噪声方差进行自适应更新，有效克服了模型噪声统计特性时变对滤波精度的影响，提升了算法的鲁棒性。最后，针对多种模态估计跟踪进行了充分的仿真分析，验证了本文方法的有效性和鲁棒性。

Hybrid System Fault Diagnosis Method Based on Noise Variance Adaptive Correction

 For hybrid system fault diagnosis problem in noise statistics properties time-varying, this paper proposed a multimodal fault diagnosis method based on noise variance adaptive relevant correction. First of all, within the framework of particle filter, hybrid system fault diagnosis is modeled as optimal state estimation and tracking problem, and real-time observation information and each modal prior transition probability is used to estimate the optimal fault mode. The estimating results are modeled separately for incoming analysis. Second, the noise variance adaptive online detection mechanism is built based on the correlation between smoothing estimation and the observation information, and updates the modal noise variance adaptively, which effectively overcomes the filter shift problem results from the time-varying noise statistical properties. The proposed method improves the robustness effectively. Finally, the experiments of three kinds failure modes show that the proposed method is efficient and robust.