等价权修正的Kalman抗野值滤波算法

Modified Kalman filter algorithm using equivalent weights with immunity to outliers

水声工程实际数据中野点的存在影响标准Kalman滤波精度。借鉴抗差惯性导航定位中等价权思想,构造反映观测值误差的统计量,以其协方差矩阵作为判断滤波精度的指标,依据莱特准则设计了三种不同权重因子,充分利用正确观测,限制利用可用观测,排除有害粗差。仿真结果表明,这种等价权修正Kalman抗差算法,抗差性可媲美中位数滤波,证明当模型正确时,其在抑制零星和斑点型野值方面,效果很好。

The equivalent weighting algorithm of robust estimation for inertial navigation and positioning is applied to the standard Kalman filter when the outliers exist in the actual data. Statistic that reflects observation error is constructed with its covariance matrix acting as the indicator of the filtering accuracy. Thus the new algorithm makes the most of the correct observations, limits the median quality observation and removes harmful gross errors. Simulation results showed that the robustness is compatible with the median filtering. When the model is correct, the modified filtering algoldthm is efficiently resistant to scattered and spotted outliers.