面向连续离散系统的自适应最大相关熵滤波算法

Adaptive Maximum Correntropy Filtering Algorithm for Continuous-Discrete Systems

针对目标跟踪中运动模型不精确和测量异常导致的传统滤波算法精度下降问题,提出了一种鲁棒平方根连续-离散自适应最大相关熵容积卡尔曼滤波(RSRCD-AMCCKF)算法。在目标跟踪问题中采用了精度更高的连续-离散时间模型,提高了目标跟踪的解算精度;将加权最小二乘方法与传统最大相关熵准则相结合,得到改进的相关熵代价权函数,之后引入连续-离散时间滤波框架,提高了滤波算法在测量异常情况下的鲁棒性;以高斯核函数作为相关熵的调整因子,依据不同测量环境选择自适应因子,进而对观测噪声的协方差矩阵进行调整。仿真结果表明:与传统算法相比,当测量噪声为高斯噪声时,RSRCD-AMCCKF算法对目标位置和速度估计的精度分别提高了38.4%和27.3%;当测量噪声为非高斯噪声时,RSRCD-AMCCKF算法对目标位置和速度估计的精度分别提高了23.5%和23.9%;当测量值发生突变时,RSRCD-AMCCKF算法对目标位置和速度估计的精度分别提高了12.6%和7.1%。RSRCD-AMCCKF算法在各类测量条件下都具有更高的精度和鲁棒性,更接近目标跟踪的克拉美罗下界,能够较好地实现滤波精度和抗异常测量的统一。

For the problems of inaccuracy of the motion model and the decrease of the traditional algorithms’filtering accuracy caused by abnormal measurement in target tracking,a robust square root continuous-discrete adaptive maximum correntropy cubature Kalman filter(RSRCD-AMCCKF)algorithm is proposed.The continuous-discrete time model with higher accuracy is adopted in target tracking,which improves the solution accuracy of target tracking;the weighted least squares method is combined with the traditional maximum correntropy criterion to obtain an improved correntropy cost weight function,and then the correntropy cost weight function is introduced into the continuous-discrete time filtering framework,which improves the robustness of the filtering algorithm under abnormal measurement;the Gaussian kernel function is used as an adjustment factor of the correntropy,and the adaptive factor is selected according to different measurement environments to adjust the observation noise covariance matrix.The simulation results show that compared with the traditional algorithms,when the measurement noise is Gaussian noise,the accuracy of the RSRCD-AMCCKF algorithm for the target position and velocity estimation is improved by 38.4%and 27.3%,respectively;when the measurement noise is non-Gaussian noise,the accuracy of the RSRCD-AMCCKF algorithm for the target position and velocity estimation is improved by 23.5%and 23.9%,respectively;when the measurement changes abruptly,the accuracy of the RSRCD-AMCCKF algorithm for the target position and velocity estimation is increased by 12.6%and 7.1%,respectively.The RSRCD-AMCCKF algorithm has higher accuracy and robustness under various measurement conditions,which is closer to the Cramer-Rao lower bound for target tracking and realizes the unification of improving filtering accuracy and suppressing abnormal measurement.