Abstract: Aiming at the problem of particle degeneracy and sample impoverishment in fault prediction, an improving particle filter with support vector regression was proposed. The no-linear function of particle state and its weight was established by using support vector regression to estimate the particle's posterior probability density model. Based on this model, the new particle was obtained and weights of particles were updated by resampling, and the diversity and effectiveness of samples were improved. Thus the ability to control and predict the fault was raised. Simulation result demonstrated that the method was feasible and system fault could be predicted correctly.

Key words: particle filter, sample improverishment, particle degeneracy, fault prediction, support vector machines

摘要： 为了对系统进行故障预测，针对粒子滤波在故障预测中出现的样本贫化与退化问题，提出了一种支持向量回归粒子滤波器。采用支持向量回归方法建立粒子状态与其权值的非线性函数来估计粒子的连续后验概率密度模型。基于该模型进行重采样获得新的粒子集并更新各粒子的权重，增加样本的多样性与有效性，提高对故障的监控与预测能力。仿真结果表明，该方法是可行的，能够准确预报系统故障。

关键词: 粒子滤波器, 样本贫化, 粒子退化, 故障预测, 支持向量机