基于地理条件对抗生成网络的车辆轨迹隐私保护方法

辛崇实; 王立勇; 冀浩杰; 金龙; 胡特

doi:10.3963/j.jssn.1674-4861.2026.01.007

基于地理条件对抗生成网络的车辆轨迹隐私保护方法

doi: 10.3963/j.jssn.1674-4861.2026.01.007

北京信息科技大学机电工程学院，北京 100192

基金项目:

北京教委科技计划项目 KM202411232005

详细信息

作者简介:
辛崇实（1999—），硕士研究生.研究方向：智能汽车与车联网信息安全. E-mail：2023020128@bistu.edu.cn

通讯作者:
冀浩杰（1988—），博士，副研究员.研究方向：智能车辆信息安全、车联网安全防护技术等. E-mail：jihaojie@bistu.edu.cn

中图分类号: TP309.2
计量
- 文章访问数: 58
- HTML全文浏览量: 17
- PDF下载量: 5
- 被引次数: 0
出版历程
- 收稿日期: 2025-08-14
- 网络出版日期: 2026-08-28

Map-conditioned Generative Adversarial Networks for Automotive Trajectory Privacy Protection

College of Mechanical and Electrical Engineering, Beijing Information Science & Technology University, Beijing 100192, China

摘要

摘要: 针对伪轨迹生成方法在大规模序列建模中梯度消失和地理语义考虑不足的轨迹失真问题，提出了地理条件约束下基于对抗生成网络的车辆轨迹数据隐私保护方法（map-conditioned trajectory generation, MCTG）。利用谱归一化生成对抗网络（spectral normalization generative adversarial networks，SN-GAN）学习全局轨迹空间分布，通过约束判别器谱范数稳定长序列训练过程，有效抑制模式崩塌与梯度消失；构建融合地图语义的条件生成对抗网络，引入编码器-解码器结构提取道路网络特征，并采用对称双向长短期记忆网络（long short-term memory，LSTM）与教师强制机制联合建模轨迹时序依赖与边界约束，从而提升生成轨迹的地理结构合理性与语义一致性。在geolife与portugal这2个真实大规模轨迹数据集上的实验结果表明：生成轨迹与真实轨迹的余弦相似度高达0.98，复杂多模态场景下稳定在0.90~0.95；詹森-香农散度（Jensen-Shannon divergence，JSD）由0.22收敛至0.105，表明生成轨迹分布性良好；而豪斯多夫距离整体低于k-匿名与差分隐私方法，空间形态保持度更高；在隐私攻击模拟实验中，假轨迹识别率低于0.01，判别准确率接近0.519，表明真假轨迹难以区分。因此，MCTG在保持轨迹统计分布一致性与道路语义约束的同时，生成的高隐私性与地理语义合理的虚假轨迹，能够有效实现车辆轨迹数据的隐私保护。
- 交通安全 /
- 车辆轨迹数据 /
- 生成模型 /
- 生成对抗网络 /
- 轨迹隐私保护
Abstract: Privacy protection methods based on pseudo-trajectory generation face vanishing gradients in large-scale sequence modeling and trajectory distortion due to insufficient modeling of geographic semantics. To address these problems, this paper proposes a map-conditioned trajectory generation framework (MCTG) based on generative adversarial networks. SN-GAN is employed to learn the global spatial distribution of trajectories. Constraints on the spectral norm of the discriminator stabilize training over long sequences and suppress mode collapse. A conditional GAN incorporating geographic semantics is further constructed, with an encoder-decoder architecture for extracting features of road networks. A symmetric bidirectional long short-term memory (LSTM) combined with a teacher-forcing mechanism jointly models temporal dependencies and boundary constraints of trajectory sequences. Experiments on the geolife and portugal datasets show that the cosine similarity between synthetic and real trajectories reaches up to 0.98, remaining stable within 0.90~0.95 under complex multi-modal conditions. The Jensen-Shannon divergence decreases from 0.22 to 0.105, indicating good distributional consistency. The Hausdorff distance of the synthetic trajectories is consistently lower than those of k-anonymity and differential privacy, reflecting superior preservation of spatial morphology. In simulated privacy attacks, the recognition rate for synthetic trajectories falls below 0.01 and the discriminator accuracy approaches 0.519, indicating that real and synthetic trajectories are nearly indistinguishable. MCTG enables effective privacy protection for vehicle trajectory data by generating synthetic trajectories with high geographic plausibility while preserving statistical distribution consistency and road-network semantic constraints.
- traffic safety /
- automotive trajectory data /
- generative models /
- generative adversarial networks /
- trajectory privacy protection

HTML全文

图 1 地图条件车辆轨迹生成隐私保护架构图

Figure 1. Map-conditioned automotive trajectory generation privacy protection architecture diagram

下载: 全尺寸图片幻灯片

图 2 真假轨迹相似性得分图

Figure 2. Similarity score chart of real and generated trajectories

下载: 全尺寸图片幻灯片

图 3 真假轨迹JSD散度收敛图

Figure 3. Convergence plot of JSD between real and generated trajectories

下载: 全尺寸图片幻灯片

图 4 豪斯多夫距离算法对比折线图

Figure 4. Comparison of Hausdorff distance among different algorithms

下载: 全尺寸图片幻灯片

图 5 波尔图地图可视化轨迹

Figure 5. Porto map visualization of trajectory

下载: 全尺寸图片幻灯片

图 6 北京地图可视化轨迹

Figure 6. Beijing map visualization of trajectory

下载: 全尺寸图片幻灯片

图 7 轨迹序列长度分布图

Figure 7. Distribution chart of trajectory sequence lengths

下载: 全尺寸图片幻灯片

图 8 真假轨迹鉴别率折线图

Figure 8. Line chart of true and false trajectory discrimination rate

下载: 全尺寸图片幻灯片

图 9 假轨迹识别率折线图

Figure 9. Line chart of fake trajectory true positive rate

下载: 全尺寸图片幻灯片

图 10 互信息值算法对比折线图

Figure 10. Comparison of mutual information among different algorithms

下载: 全尺寸图片幻灯片

参考文献(30)

[1]	李克强, 李家文, 常雪阳, 等. 智能网联汽车云控系统原理及其典型应用[J]. 汽车安全与节能学报, 2020, 11(3): 261-275. LI K Q, LI J W, CHANG X Y, et al. Principles and typical applications of intelligent connected vehicle cloud control system[J]. Journal of Automotive Safety and Engergy, 2020, 11 (3): 261-275. (in Chinese)
[2]	奇格奇, 曹琳琪, 沈益达, 等. 基于GNSS轨迹数据的公交多能源供需网络调度优化模型[J]. 交通信息与安全, 2025, 43 (3): 85-99. QI G Q, CAO L Q, SHEN Y D, et al. Optimization model for multi-energy supply and demand network scheduling of public transport based on GNSS trajectory data[J]. Journal of Transport Information and Safety, 2025, 43(3): 85-99. (in Chinese)
[3]	胥川, 胡家琳, 宫丽婷, 等. 考虑交通安全的车载导航技术研究进展[J]. 交通信息与安全, 2025, 43(6): 1-10, 20. XU C, HU J L, GONG L T, et al. Research progress of vehicle navigation technology considering traffic safety[J]. Journal of Transport Information and Safety, 2025, 43(6): 1-10, 20. (in Chinese)
[4]	ZHANG S, LI M, LIANG W, et al. A survey of dummy-based location privacy protection techniques for location-based services[J]. Sensors, 2022, 22(16): 6141. doi: 10.3390/s22166141
[5]	OUYANG Y, XU Y, LE Z, et al. Providing location privacy in assisted living environments[C]. The 1st International Conference on Pervasive Technologies Related to Assistive Environments, New York, United States: Association for Computing Machinery, 2008.
[6]	SWEENEY L. K-anonymity: a model for protecting privacy[J]. International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems, 2002, 10(5): 557-570. doi: 10.1142/S0218488502001648
[7]	邓劲松, 王传安, 吴延敏. 基于抑制和敏感属性多样性的轨迹发布算法[J]. 阜阳师范大学学报(自然科学版), 2022, 39 (2): 91-98. DENG J S, WANG C A, WU Y M. Trajectory publishing algorithm based on inhibition and sensitivity attribute diversity[J]. Journal of Fuyang Normal University (Natural Science), 2022, 39(2): 91-98. (in Chinese)
[8]	CHEN R, FUNG B C M, DESAI B C, et al. Differentially private transit data publication: a case study on the montreal transportation system[C]. The 18^th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, New York, United States: Association for Computing Machinery, 2012.
[9]	ZHOU B, PEI J. The k-anonymity and l-diversity approaches for privacy preservation in social networks against neighborhood attacks[J]. Knowledge and Information Systems, 2011, 28(1): 47-77. doi: 10.1007/s10115-010-0311-2
[10]	LI N, LI T, VENKATASUBRAMANIAN S. T-closeness: privacy beyond k-anonymity and l-diversity[C]. 2007 IEEE 23^rd International Conference on Data Engineering Workshop, Istanbul, Turkey: IEEE, 2006.
[11]	许华杰, 吴青华, 胡小明. 基于轨迹多特性的隐私保护算法[J]. 计算机科学, 2019, 46(1): 190-195. XU H J, WU Q H, HU X M. Privacy protection algorithm based on multiple trajectory features[J]. Computer Science, 2019, 46(1): 190-195. (in Chinese)
[12]	钟小宇, 李梦涵, 李丽红. 车联网位置隐私保护研究综述[J]. 物联网技术, 2023, 13(3): 77-79. ZHONG X Y, LI M H, LI L H. A survey on location privacy protection in the internet of vehicles[J]. Internet of Things Technology, 2023, 13(3): 77-79. (in Chinese)
[13]	LI J, CHEN W, LIU A, et al. FTS: a feature-preserving trajectory synthesis model[J]. Geoinformatica, 2018, 22: 49-70. doi: 10.1007/s10707-017-0301-6
[14]	YU Y, ZHU H, XIE M. CTP: correlated trajectory publication with differential privacy[C]. The 6^th International Conference on Computer and Communication Systems, Cheng du, China: IEEE, 2021.
[15]	樊娜, 罗闯, 张泽晖, 等. 基于改进生成对抗网络的车辆轨迹语义隐私保护机制[J]. 计算机应用, 2026, 46(1): 169-180. FAN N, LUO C, ZHANG Z H, et al. Vehicle trajectory semantic privacy protection mechanism based on improved generative adversarial networks[J]. Computer Applications, 2026, 46(1): 169-180. (in Chinese)
[16]	王超, 张泽晖, 樊娜, 等. 深度生成模型的轨迹数据发布隐私保护机制[J]. 交通运输工程学报, 2025, 25(4): 340-354. WANG C, ZHANG Z H, FAN N, et al. Privacy protection mechanism for publishing trajectory data of deep generative models[J]. Journal of Traffic and Transportation Engineering, 2025, 25(4): 340-354. (in Chinese)
[17]	BARATCHI M, MERATNIA N, HAVINGA P J M, et al. A hierarchical hidden semi-markov model for modeling mobility data[C]. The 2014 ACM International Joint Conference on Pervasive and Ubiquitous Computing, New York, United States: Association for Computing Machinery, 2014.
[18]	姚尧, 蒋应红, 邹国建, 等. 路网约束下基于动态速度模拟的个体轨迹重构[J]. 交通运输工程与信息学报, 2026, 24 (1): 38-49. YAO Y, JIANG Y H, ZOU G J, et al. Individual trajectory re-construction based on dynamic velocity simulation under road network constraints[J]. Journal of Transportation Engineering and Information, 2026, 24(1): 38-49. (in Chinese)
[19]	GULRAJANI I, AHMED F, ARJOVSKY M, et al. Improved training of wasserstein gans[C]. 31^st Annual Conference on Neural Information Processing Systems, California, United States: NeurlPS, 2017.
[20]	SHU Z, ZHANG K. Spectral normalization for generative adversarial networks for artistic image transformation[J]. International Journal of Digital Multimedia Broadcasting, 2024 (1): 6644706.
[21]	GARG S, PEITZ S, NALLASAMY U, et al. Jointly learning to align and translate with transformer models[C]. The 2019 Conference on Empirical Methods in Natural Language Processing and The 9^th International Joint Conference on Natural Language Processing, Hong Kong, China: Association for Computational Linguistics, 2019.
[22]	LI Y, CHEN C, DUAN M, et al. Attention-aware encode-decoder neural networks for heterogeneous graphs of things[J]. IEEE Transactions on Industrial Informatics, 2020, 17(4): 2890-2898.
[23]	李康, 于娟, 韩建民, 等. 融合多源城市环境信息的知识图谱驱动轨迹生成模型[J]. 电子学报, 2025, 53(10): 3551-3565. LI K, YU J, HAN J M, et al. Knowledge graph-driven trajectory generation model that integrates multi-source urban environmental information[J]. Acta Electronica Sinica, 2025, 53 (10): 3551-3565. (in Chinese)
[24]	SHEN Y, HUA R, WANG H, et al. Personalized trajectory differential privacy protection mechanism based on spatiotemporal correlation prediction[J]. Academic Journal of Computing & Information Science, 2024, 7(1): 1-8.
[25]	SHEN H, WANG Y, ZHANG M. A privacy-preserving trajectory publishing method based on multi-dimensional sub-trajectory similarities[J]. Sensors, 2023, 23(24): 9652. doi: 10.3390/s23249652
[26]	YANG X, ZHANG C. A location trajectory privacy protection method based on generative adversarial network and attention mechanism[J]. Computers, Materials & Continua, 2024, 81(3): 3781.
[27]	WANG X, LIU X, Lu Z, et al. Large scale GPS trajectory generation using map based on two stage GAN[J]. Journal of Data Science, 2021, 19(1): 126-141.
[28]	WANG Y, LI M, LUO S, et al. LRM: a location recombination mechanism for achieving trajectory k-anonymity privacy protection[J]. IEEE Access, 2019(7): 182886-182905.
[29]	CHEN S, FU A, SHEN J, et al. RNN-DP: a new differential privacy scheme base on recurrent neural network for dynamic trajectory privacy protection[J]. Journal of Network and Computer Applications, 2020, 168: 102736. doi: 10.1016/j.jnca.2020.102736
[30]	金朋竺. 车联网中轨迹隐私保护技术研究[D]. 绵阳: 西南科技大学, 2024. JIN P Z. Research on trajectory privacy protection technologies in the internet of vehicles[D]. Mianyang: Southwest University of Science and Technology, 2024. (in Chinese)