基于遗传粒子滤波器的传感器故障诊断

Sensor fault diagnosis based on genetic particle filter

针对标准粒子滤波器(PF)中存在的粒子退化及计算量过大问题,将遗传算法中的选择、交叉、变异操作引入到PF中,代替传统的重采样方法,提出了一种改进的遗传粒子滤波器(GPF)算法。算法采用实数运算,从而避免了二进制编码操作。应用GPF与PF诊断三种常见的恒偏差、恒增益、卡死传感器故障,并比较了两者的诊断效果。仿真结果表明,GPF增加了粒子多样性,能有效解决粒子退化问题,提高滤波精度和实时性。

To solve the problems of particle degeneracy and calculation overload in particle filter（PF）,selection,cross-over and mutation operation in the genetic algorithm were applied to PF to replace the traditional resampling method,such that an improved genetic particle filter（GPF）algorithm is proposed,which uses real number operation instead of binary coding.GPF and PF are adopted to diagnose three types of common sensor faults,i.e.,constant deviation,constant gain and seizing-up.The diagnosis results of these two filters are compared with each other.Simulation results show that GPF can increase particle diversity,solve the problem of particle degeneracy effectively,and improve filtering accuracy and real time performance.