基于粒子滤波的陀螺随机漂移预测模型研究

Particle filter based modeling research for gyroscope random drift

陀螺随机漂移是弱的非平稳非线性信号,传统的线性化参数估计方法会导致模型参数不精确。提出采用粒子滤波的方法对陀螺随机漂移时间序列AR模型进行滤波处理,试验结果表明:粒子滤波不仅消除了非线性对陀螺漂移的影响还克服了AR模型的参数估计误差。与UKF滤波方法相比,粒子滤波的随机量不受高斯分布的限制而且滤波结果可以达到或超过UKF。

Gyroscope random drift is a weak, non-linear and non-stationary signal. Traditional linear parameter estimation methods may lead to inaccurate model parameters. This paper adopts particle filter to process the established time series AR model of gyroscope random drift. Experiment result shows that particle filter not only gets rid of the impacts of the non-linear gyroscope drift, but also wipes off the AR model parameter estimation errors. As a conclusion, particle filter random quantity isn’t restricted by Gaussian distribution and can reach or execeed the results of UKF.