Research on vehicle integrated navigation system based on improved sample convolutional interaction network

doi:10.3969/j.issn.1674-8484.2025.02.015

Abstract

Abstract:

Global Navigation Satellite System / Inertial Navigation System (GNSS/INS) integrated navigation system in vehicles is prone to signal loss in obstructed environments, leading to divergent positioning results and compromising the efficiency and safety of unmanned vehicles. To address this issue, this study proposed an artificial intelligence solution based on an improved Sample Convolution and Interaction Network (SCINet), which incorporated strategies such as principal component analysis, trend decomposition, and linear convolutional interactive learning on a low-layer SCINet architecture, enhancing the stability and accuracy of the model under such operating conditions. The results show that the proposed model reduces positioning errors by 80.9% and 67.6% compared to Long Short-Term Memory (LSTM) and SCINet, respectively, effectively improving the outdoor positioning accuracy of unmanned vehicles during GNSS signal loss and ensuring the reliability and safety of unmanned vehicle positioning.

Key words: unmanned vehicles, integrated navigation, inertial navigation system (INS) outage, sample convolutional interaction network (SCINet), trend seasonal separating

CLC Number:

TP183

KUANG Xinghong, YAN Biyun. Research on vehicle integrated navigation system based on improved sample convolutional interaction network[J]. Journal of Automotive Safety and Energy, 2025, 16(2): 315-325.

Figures/Tables 19

References 16

[1]	ZHAO Zhongyi, CUI Yongzhi. State measurement and control technology of unmanned vehicle based on high-precision positioning[J]. Mobi Info Syst, 2021, 2021(6): 1-9.
[2]	ZHOU Jingmei, ZHAO Xiangmo, CHENG Xin et al. Vehicle ego-localization based on streetscape image database under blind area of global positioning system[J]. *J Shanghai Jiaotong University* (Sci Edit)**, 2019, 24(1): 122-129
[3]	王可, 刘立刚, 周斌, 等. 基于SINS/GNSS/轮速的列车组合导航算法研究[J]. 系统仿真学报, 2021, 33(1): 189-195. doi: 10.16182/j.issn1004731x.joss.19-0188
	WANG Ke, LIU Ligang, ZHOU Bin, et al. Research on SINS/GNSS/train-speed integrated navigation algorithm[J]. J Syst Simulat, 2021, 33(1): 189-195. (in Chinese)
[4]	章红平, 高周正, 牛小骥, 等. GPS非差非组合精密单点定位算法研究[J]. 武汉大学学报 (信息科学版), 2013, 38(12): 1396-1399
	ZGANG Hongping, GAO Zhouzheng, NIU Xiaoji, et al. Research on GPS precise point positioning with un-differentialand un-combined observations[J]. Geom Info Sci Wuhan Univ, 2013, 38(12): 1396-1399. (in Chinese)
[5]	XIAO Xuan, SUN Jiaxing, ZHANG Qin. A fusion method based on ANN to overcome the GNSS Outages for GNSS/INS System [C]// 2021 China Autom Congress (CAC). Beijing: Chinese Association of Automation, 2021: 4914-4919
[6]	MENG Xiaoliang, TAN Hongbin, YAN Peihui et al. A GNSS/INS integrated navigation compensation method based on CNN-GRU + IRAKF hybrid model during GNSS outages[J]. IEEE Trans Instrum Measurem, 2024, 73: 1-15.
[7]	刘东亮, 成芳, 沈朋礼, 等. LSTM辅助车载GNSS/INS组合导航算法及性能分析[J]. 全球定位系统, 2023, 48(5): 21-31.
	LIU Dongliang, CHENG Fang, SHEN Pengli, et al. LSTM assisted in vehicle GNSS/INS integrated navigation algorithm and performance analysis[J]. GNSS World China, 2023, 48(5): 21-31. (in Chinese)
[8]	ZHANG Yuexin, WANG Lihui. A hybrid intelligent algorithm DGP-MLP for GNSS/INS integration during GNSS outages[J]. J Navigat, 2019, 72(2): 375-388.
[9]	LIU Minhao, ZENG Ailing, LAI Qiuxiai, et al. Time series is a special sequence: Forecasting with sample convolution and interaction[J/OL]. 36th Conf Neur Info Proc Syst (NeurIPS 2022). 2021. (2021-06-17) https://doi.org/10.48550/arXiv.2106.09305.
[10]	Hamilton J D. Time Series Analysis [M]. Princeton, New Jersey, USA: Princeton University Press, 2020: 15-31.
[11]	胡小曼, 王艳, 纪志成. 模糊信息粒化与改进RVM的滚动轴承寿命预测[J]. 系统仿真学报, 2021, 33(11): 2561-2571.
	HU Xiaoman, WANG Yan, JI Zhicheng. Fuzzy information granulation and improved RVM for rolling bearing life prediction[J]. J Syst Simulat, 2021, 33(11): 2561-2571. (in Chinese)
[12]	陈海燕, 欧阳林威, 任宝民. 一种基于频域注意力机制变量筛选的多变量时间序列预测模型[J/OL]. 华中科技大学学报(自然科学版), 2024(3):1-10. [2024-04-26] http://doi.org/10.13245/j.hust.250812.
	CHEN Haiyan, OUYANG Linwei, REN Baomin, et al. Multivariate time series forecasting for screening important variables based on frequency domain attention mechanism[J/OL]. J Huazhong Univ Sci Tech (Nat Sci Edit), 2024(3): 1-10. [2024-04-26] http://doi.org/10.13245/j.hust.250812. (in Chinese)
[13]	WU Haixu, XU Jiehui, WANG Jianmin, et al. Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting[J]. Neur Info Proc Syst, 2021(3): 22419-22430.
[14]	WANG Shiyu, WU Haixu, SHI Xiaoming, et al. TIMEMIXER: Decomposable multiscale mixing for time series forecasting[J/OL]. arXiv 2024: 2405.14616V1 [cs.LG]. [2024-07-21] https://doi.org/10.48550/arXiv.2405.14616.
[15]	Oreshkin B N, Carpov D, Chapados N, et al. N-BEATS: Neural basis expansion analysis for interpretable time series forecasting [J/OL]. arXiv 2019: 1905.10437V4 [cs.LG]. [2024-07-25] https://doi.org/10.48550/arXiv.1905.10437.
[16]	ZENG Ailing, CHEN Muxi, ZHANG Lei, et al. Are transformers effective for time series forecasting?[C]// Proc AAAI Conf Artif Intel. Washington DC, USA. 2023, 37(9): 11121-11128.

参数	精度
陀螺仪静止零漂	10 (°) / h
陀螺仪噪声	$60\left(^{\circ}\right) /\left[\mathrm{s}(\sqrt{\mathrm{~Hz}})^{-1}\right]$
陀螺仪采样频率	100 Hz
加速度计静止零漂	(±15) μg
加速度计噪声	60 μg / Hz^-2
加速度计采样频率	100 Hz
GNSS精度(水平)	0~2 m
GNSS采样频率	5 Hz

参数	精度
陀螺仪静止零漂	10 (°) / h
陀螺仪噪声	$60\left(^{\circ}\right) /\left[\mathrm{s}(\sqrt{\mathrm{~Hz}})^{-1}\right]$
陀螺仪采样频率	100 Hz
加速度计静止零漂	(±15) μg
加速度计噪声	60 μg / Hz^-2
加速度计采样频率	100 Hz
GNSS精度(水平)	0~2 m
GNSS采样频率	5 Hz

参数	精度
陀螺仪静止零漂	±(0.5~1) (°) / s
陀螺仪RMS噪声	0.028~0.07 (°) / s
陀螺仪分辨率 (LSB)	0.06 (°) / s
陀螺仪采样频率	50 Hz
加速度计量程	(±16) g
加速度计静止零漂	±(20~40) mg
加速度计RMS噪声	0.75~1 mg/Hz
加速度计采样频率	50 Hz
GNSS精度	水平2.5 m
GNSS采样频率	1 Hz

参数	精度
陀螺仪静止零漂	±(0.5~1) (°) / s
陀螺仪RMS噪声	0.028~0.07 (°) / s
陀螺仪分辨率 (LSB)	0.06 (°) / s
陀螺仪采样频率	50 Hz
加速度计量程	(±16) g
加速度计静止零漂	±(20~40) mg
加速度计RMS噪声	0.75~1 mg/Hz
加速度计采样频率	50 Hz
GNSS精度	水平2.5 m
GNSS采样频率	1 Hz

模型	FLOPs/次	参数量	t_mean / ms
D-SCINet	5 491 200	28 976	19.2
SCINet-High-Level	11 625 984	73 172	69.8
SCINet-Low-Level	3 783 168	10 700	9.9