基于混合熵优化下的卡尔曼滤波组合导航算法OA
Optimized Kalman filtering approach for integrated navigation based on entropy fusion
为解决传统卡尔曼滤波器在时变厚尾非高斯噪声环境下的性能退化问题,提出了一种基于分布熵(Distribution entropy,DE)的最大相关熵卡尔曼滤波器(Maximum correntropy Kalman filter,MCKF)算法,即DE-MCKF.该算法利用分布熵来衡量测量数据误差的混淆程度,并结合自适应变化策略,在保持MCKF滤波效果的同时,优化其在不同非高斯噪声环境下的性能.当熵值较高导致过度变化时,通过分配熵值可及时提升对测量值的信任度,从而避免过度依赖模型预测,减少因MCKF对噪声误判而引起的滤波误差.实验结果表明,与KF、DEKF、MCKF和VBKF相比,DE-MCKF分别将平均纬度和经度误差的均方根误差降低了62.76%、58.46%、55.11%和43.17%,有效提升了系统在非高斯噪声环境下的定位精度,为高精度组合导航及鲁棒状态估计提供了一种新的方案.
To address the performance degradation of traditional Kalman filters in time-varying thick-tailed non-Gaussian noise environments,we propose a maximum correntropy Kalman filter(MCKF)algorithm based on distribution entropy(DE),called DE-MCKF.Among them,the DE is used to determine the confusion degree of measurement data errors,combined with the adaptive change strategy to optimize the MCKF under varied non-Gaussian noise while maintaining the filtering effect of the MCKF.Distributing the entropy can timely improve trust in the measurement value when a high entropy value causes over-change,avoiding over-reliance on model predictions and reducing the filtering errors from MCKF misjudgments on the noise.The results show that DE-MCKF reduces the root mean squared error(RMSE)of averaged latitude and longitude error by 62.76%,58.46%,55.11%,and 43.17%compared to KF,DEKF,MCKF and VBKF,respectively.It can effectively improve the positioning accuracy of the system under non-Gaussian noise.
张翔;周鑫;司涌波;徐万镛;陈光武
兰州交通大学 自动化与电气工程学院,甘肃 兰州 730070兰州交通大学 自动化与电气工程学院,甘肃 兰州 730070兰州交通大学 自动化与电气工程学院,甘肃 兰州 730070兰州交通大学 自动化与电气工程学院,甘肃 兰州 730070兰州交通大学 自动化与电气工程学院,甘肃 兰州 730070
最大熵分布熵卡尔曼滤波非高斯噪声高斯核函数
maximum correntropy criterion(MCC)distribution entropy(DE)Kalman filternon-Gaussian noiseGaussian kernel function
《测试科学与仪器》 2026 (3)
384-394,11
This study was supported in part by National Natural Science Foundation of China(No.52472344),Gansu Provincial Department of Education(No.2023CYZC-32),Gansu Institute of Integrated Circuit Industry(No.[2023]36),and Lanzhou Science and Technology Program(No.2025-GN-1).
评论