陈永刚, 王妍, 白邓宇, 熊文祥. 基于LSTM网络辅助无迹粒子滤波的列车定位方法研究[J]. 云南大学学报(自然科学版), 2021, 43(3): 477-485. doi: 10.7540/j.ynu.20200388
引用本文: 陈永刚, 王妍, 白邓宇, 熊文祥. 基于LSTM网络辅助无迹粒子滤波的列车定位方法研究[J]. 云南大学学报(自然科学版), 2021, 43(3): 477-485. doi: 10.7540/j.ynu.20200388
CHEN Yong-gang, WANG Yan, BAI Deng-yu, XIONG Wen-xiang. Research on train positioning method based on LSTM network aided unscented particle filter[J]. Journal of Yunnan University: Natural Sciences Edition, 2021, 43(3): 477-485. DOI: 10.7540/j.ynu.20200388
Citation: CHEN Yong-gang, WANG Yan, BAI Deng-yu, XIONG Wen-xiang. Research on train positioning method based on LSTM network aided unscented particle filter[J]. Journal of Yunnan University: Natural Sciences Edition, 2021, 43(3): 477-485. DOI: 10.7540/j.ynu.20200388

基于LSTM网络辅助无迹粒子滤波的列车定位方法研究

Research on train positioning method based on LSTM network aided unscented particle filter

  • 摘要: 针对列车在实际运行过程中,在通视性不良区域难以获得全球定位系统(Global Positioning System,GPS)信号,致使GPS/捷联惯性导航系统(Strapdown Intertial Navigation System,SINS)列车定位系统精度下降的情况,提出一种基于长短时记忆(Long Short Term Memory,LSTM)网络无迹粒子滤波(Unscented Particle Filter,UPF)的定位方法. 在GPS信号有效的情况下,使用UPF1进行列车定位,并利用UPF1输出的位置速度信息训练LSTM1;当GPS信号缺失时,引入神经网络监督控制思想,使用训练好的LSTM1代替GPS信号,并将其与SINS输出信息作为反馈控制器UPF2的输入,使用UPF2的输入输出搭建神经网络控制器LSTM2;系统的输出由UPF2和LSTM2的输出共同决定,但随着LSTM2不断逼近系统模型,会取代UPF2决定最终输出结果. 仿真结果证明,采用LSTM辅助UPF的方法可以满足列车定位的要求.

     

    Abstract: Aiming at the situation that the accuracy of GPS/SINS train positioning system is reduced due to the difficulty of obtaining GPS signals in poor visibility areas during the actual operation of the train, a long short term memory network (LSTM) aided unscented particle filter (UPF) positioning method is proposed. When the GPS signal is available, UPF1 is used for train positioning, and the position and speed information output by UPF1 is used to train LSTM1. When the GPS signal is unavailable, the neural network supervisory control idea is introduced, and the trained LSTM1 is used instead of the GPS signal. With SINS output information as the input of the feedback controller UPF2, the neural network controller LSTM2 is built using the input and output of UPF2. The output of the system is jointly determined by the output of UPF2 and LSTM2, but as LSTM2 continues to approach the system model, it will replace UPF2 to decide the final output the result. The simulation results prove that the LSTM network aided UPF method can meet the requirements of train positioning.

     

/

返回文章
返回