一种基于改进RBF神经网络的组合导航方法  

An integrated navigation method based on improved RBF neural network

作　　者：文家燕[1,2] 张锱强 李克强[3] 宾仕博 何逸波 WEN Jiayan;ZHANG Ziqiang;LI Keqiang;BIN Shibo;HE Yibo(School of Automation,Guangxi University of Science and Technology,Liuzhou 545616,China;Research Center for Intelligent Collaboration and Cross-application,Guangxi University of Science and Technology,Liuzhou 545616,China;School of Vehicle and Mobility,Tsinghua University,Beijing 100084,China;Dongfeng Liuzhou Automobile Co.,Ltd.,Liuzhou 545005,China;SAIC GM Wuling Co.,Ltd.,Liuzhou 545007,China)

机构地区：[1]广西科技大学自动化学院,柳州545616 [2]广西科技大学智能协同与交叉应用研究中心,柳州545616 [3]清华大学车辆与运载学院,北京100084 [4]东风柳州汽车有限公司,柳州545005 [5]上汽通用五菱股份有限公司,柳州545007

出　　处：《航天控制》2025年第2期40-48,共9页Aerospace Control

基　　金：广西科技重大专项(桂科AA24206054,桂科AA24206047);国家自然科学基金项目(61963006);广西自然科学基金面上项目(2018GXNSFAA050029,2018GXNSFAA294085);广西重点研发计划(桂科AB23075093,桂科AB22035066);2022年广西汽车零部件与整车技术重点实验室自主研究课题(2022GKLACVTZZ01)。

摘　　要：针对车辆在隧道、城市路段和峡谷等复杂环境中全球导航定位系统(GNSS)信号中断导致组合导航定位精度下降的问题,提出一种基于改进径向基函数神经网络(RBF)辅助容积卡尔曼滤波(CKF)的组合导航定位方法。首先,通过核主成分分析(KPCA)结合K-means++聚类模型对组合导航融合数据进行预处理,使其分布具有代表性;其次,利用正交最小二乘法(OLS)确定RBF神经网络隐含层神经元的数量及中心值,并采用信赖域约束高斯-牛顿(TR-CGN)算法优化其参数;最后,在GNSS信号失锁时,利用训练好的改进RBF神经网络辅助非线性CKF滤波进行误差补偿。实验结果表明,该方法在不增加硬件成本的情况下,平均定位误差较自动驾驶协同定位系统降低了17.87%;与KPCA-RBF辅助的平均定位误差相比降低了54.37%,可见,所提方法有效增强了组合导航定位系统在复杂环境下的适应性和鲁棒性。A combined navigation positioning method based on improved radial basis function neural network(RBF)assisted volume Kalman filter(CKF)is proposed to resolve reduced precision of integrated navigation positioning caused by global navigation positioning system(GNSS)signal interruption in complex environments such as tunnels,urban roads and canyons.Firstly,the integrated navigation fusion data is preprocessed by using kernel principal component analysis(KPCA)combined with K-means++clustering model to make its distribution representative;Secondly,the orthogonal least squares(OLS)method is used to determine the number and center values of hidden layer neurons in the RBF neural network,and the trust region constrained Gaussian-Newton(TR-CGN)algorithm is used to optimize its parameters;Finally,when the GNSS signal loses lock,the trained improved RBF neural network is used to assist in nonlinear CKF filtering for error compensation.The experimental results show that the average positioning error is reduced by 17.87%through application of this method without increasing hardware costs which is compared to the way of using the autonomous driving collaborative positioning system;Compared with the average positioning error assisted by KPCA-RBF,the reduction based on the proposed method takes advantage of 54.37%,which indicates that the adaptability and robustness of the integrated navigation positioning system are effectively enhanced within complex environments.

关 键 词：组合导航 改进RBF OLS算法 TR-CGN算法 容积卡尔曼滤波 

分 类 号：V249.31[航空宇航科学与技术—飞行器设计] TP183[自动化与计算机技术—控制理论与控制工程]