Fault diagnosis of train sensors based on evolutionary genetic Particle Filter

Particle Filter can be used to fault diagnosis on systems with nonlinearities or non-Gaussian noise as a state estimation algorithm. Due to its characteristics to handle with discrete and continuous states simultaneously, particle filter has attracted much more attention to fault diagnosis on hybrid systems. Rao-Blackwellized Particle Filter (RBPF) is one of the efficient methods to this application without the limitation of high dimensional state spaces. However, in the implementation of particle filter, a resampling scheme is often used to mitigate the degeneracy phenomenon; meanwhile it comes out another particle deprivation problem and diversity decreased. In order to overcome this inherent problem of particle filter, an evolutionary Genetic Algorithm (EGA) integrated with RBPF is proposed, and applied to diagnose failures in hybrid train sensor system. Simulations demonstrate that the improved algorithm can significantly increase particle diversity and reduce the error rate of fault diagnosis.