Liu Fengyu, Sun Xiaohong, Xiong Yufeng, Huang Haoqian, Guo Xiaoting, Zhang Yu, Shen Chong
Key Laboratory of Instrumentation Science and Dynamic Measurement, Ministry of Education, North University of China, Taiyuan 030051, People's Republic of China.
School of Electronic and Information Engineering, Suzhou University of Science and Technology, Suzhou 215000, People's Republic of China.
Rev Sci Instrum. 2019 Dec 1;90(12):125005. doi: 10.1063/1.5094559.
To improve the performance of the Global Positioning System/Inertial Navigation System (GPS/INS) integrated navigation system, current research studies merely combine neural networks with nonlinear filter methods. Few studies focus on how to optimize the parameters of the neural network and how to further improve the small error accumulated into the next filter step due to the imprecise design of the filter when setting the initial parameters in the GPS/INS integrated system. In this article, a dual optimization method consisting of an iterated cubature Kalman filter-Feedforward Neural Network (ICKF-FNN) and a radial basis function-cubature Kalman filter (RBF-CKF) is proposed to compensate the position and velocity errors of the integrated system during GPS outages. The prominent advantages of the proposed method include the following. (i) The ICKF is designed to optimize the parameters of the introduced FNN adaptively and obtain an appropriate internal structure when GPS is available, which improves the accuracy of the training model. (ii) The RBF establishes the relationship between filter parameters and the optimal estimation errors, reducing the errors caused by inaccurate predicted observation during GPS outages. (iii) The proposed dual optimization method takes advantages over other combination algorithms under different moving conditions or even during long period of GPS outages, which shows its great stability. Experimental results show that the root mean squared error of the east position is reduced by 85.79% to 3.2187 m using the proposed strategy during turning movement and the east velocity error accumulation rate decreases by 92.69% during the long straight movement of 250 s. These results are from offline processing.
为提高全球定位系统/惯性导航系统（GPS/INS）组合导航系统的性能，当前的研究仅将神经网络与非线性滤波方法相结合。很少有研究关注如何优化神经网络的参数，以及如何在GPS/INS组合系统中设置初始参数时，因滤波器设计不精确而进一步改善在下一个滤波步骤中累积的小误差。在本文中，提出了一种由迭代容积卡尔曼滤波器-前馈神经网络（ICKF-FNN）和径向基函数-容积卡尔曼滤波器（RBF-CKF）组成的双重优化方法，以补偿GPS中断期间组合系统的位置和速度误差。该方法的显著优点包括以下几点。（i）ICKF旨在在GPS可用时自适应地优化引入的FNN的参数，并获得合适的内部结构，从而提高训练模型的精度。（ii）RBF建立了滤波器参数与最优估计误差之间的关系，减少了GPS中断期间预测观测不准确所导致的误差。（iii）所提出的双重优化方法在不同移动条件下甚至在长时间GPS中断期间优于其他组合算法，显示出其强大的稳定性。实验结果表明，在转弯运动期间，使用所提出的策略，东向位置的均方根误差降低了85.79%，降至3.2187 m，在250 s的长直线运动期间，东向速度误差累积率降低了92.69%。这些结果来自离线处理。
