基于自学习采样粒子滤波器的不完备故障空间交互诊断方法

Interactive diagnosis in incomplete fault space based on self-learning sample particle filter

针对不完备空间混合系统,提出一种基于自学习采样粒子滤波器(SLSPF)的交互诊断方法,融入自学习采样机制,利用自学习即时概率指导采样,以摆脱粒子滤波器对转移概率的依赖;结合自学习采样与诊断的动态交互方式调整模式空间,使粒子滤波器采样粒子数动态减少;同时给出了不完备信息空间的真实模式与未知模式阈值的决策条件,实验结果表明,尤其在高维状态空间下,SLSPF不仅可以保证粒子滤波器的诊断精度,而且能够提高计算效率。

To solve the problem of hybrid system fault diagnosis in incomplete space, a dynamical fault diagnosis method based on self-learning sample particle filter（SLSPF） is presented. With the mechanism of self-learning sampling and real-time distribution probability directed sampling, SLSPF can break out of the dependence on transition probability. The combination of self-learning sampling and dynamic interactive diagnosis mode makes the sampling number of the filter tend to decrease dynamically and adjusts the mode space. The threshold value decision-making condition of real mode and unknown mode in the incomplete information space is given. Experiment results show that even if in the higher-dimensional space, SLSPF can still guarantee the particle filter diagnose precision and computational efficiency.