Abstract:

In order to solve the problem of low fault diagnosis accuracy caused by Particle Filter（PF） sample depletion in the process of fault diagnosis of nonlinear and non-Gaussian complex dynamic systems, a new particle filter algorithm based on improved bird swarm optimization algorithm is proposed in this paper. Aiming at the problem that the standard bird algorithm is easy to fall into the local optimum, the dynamic self-adaptive coefficient and self-adaptive steps are introduced to introduce the position and global optimal position information of each bird into the adaptive change control, so as to improve the problem of falling into the local optimum. The improved bird swarm algorithm is adopted to optimize the particle filter resampling process, which is to imitate the bird’s foraging, warning and flight behavior to move the particles to the high likelihood region. The effectiveness of the algorithm is verified by simulation analysis of fault diagnosis of stator current sensor of doubly-fed generator. Experimental results show that this algorithm can effectively improve the accuracy of fault diagnosis.

Key words: Particle Filter（PF）, fault diagnosis, doubly-fed generator, bird swarm algorithm, self-adaptive coefficient, self-adaptive steps

摘要：

针对粒子滤波（PF）在处理非线性、非高斯复杂动态系统故障诊断过程中，由于样本贫化所导致的故障诊断准确度低的问题，提出了一种改进鸟群算法优化粒子滤波的新算法。针对标准鸟群算法容易陷入局部最优问题，引入动态自适应系数和自适应步长，把每只鸟的位置和全局最优位置信息引入到自适应变化控制中，从而改善陷入局部最优的问题；采用改进后的鸟群算法优化粒子滤波重采样过程，即通过模拟鸟群的觅食、警戒和飞行行为使得粒子移向高似然区域；通过对双馈发电机定子电流传感器故障诊断的仿真分析，验证了算法的有效性。实验结果表明此算法可有效提高故障诊断的准确度。

关键词: 粒子滤波, 故障诊断, 双馈发电机, 鸟群算法, 自适应系数, 自适应步长