ICLR 2026 | 时空数据(Spatial-Temporal)论文总结[上]（交通与城市科学：交通预测，轨迹挖掘，自动驾驶等）

ICLR 2026将在2025年4月24日到28日于巴西里约热内卢（Rio de Janeiro, Brazil）举行。ICLR 2025共有19,000多篇投稿，录用5,359篇，录取率28.18%。本文总结了2026 ICLR上有关时间序列（time series）相关论文。如有疏漏，欢迎大家补充。

本文总结了ICLR 2026时空数据（Spatial-Temporal）的论文，总计36 篇，本文涉及23篇，如有疏漏，欢迎补充。

注 ：由于论文数目较多，分为上下篇 ，基于数据生成机制与应用场景的本质差异对论文进行分类：

• 上篇 主要涵盖交通时空数据以及城市科学等内容，包括交通预测，人群移动，轨迹挖掘，交通模拟，自动驾驶，信号控制等内容。

• 下篇 主要涵盖：气象时空和物理时空，如气象预测，时空点过程，时空动力系统等

观察：上篇文章统计值

最大均分	均值	最小均分
6.5	5.28	4

其中 均分≥6的有6篇

1. Micro-Macro Coupled Koopman Modeling on Graph for Traffic Flow Prediction 2. MoRA: Mobility as the Backbone for Geospatial Representation Learning at Scale 3. ELLMob: Event-Driven Human Mobility Generation with Self-Aligned LLM Framework 4. Lightweight Spatio-Temporal Modeling via Temporally Shifted Distillation for Real-Time Accident Anticipation 5. ST-HHOL: Spatio-Temporal Hierarchical Hypergraph Online Learning for Crime Prediction 6. A General Spatio-Temporal Backbone with Scalable Contextual Pattern Bank for Urban Continual Forecasting 7. USTBench: Benchmarking and Dissecting Spatiotemporal Reasoning Capabilities of LLMs as Urban Agents 8. UrbanGraph: Physics-Informed Spatio-Temporal Dynamic Heterogeneous Graphs for Urban Microclimate Prediction 9. TrajFlow: Nation-wide Pseudo GPS Trajectory Generation with Flow Matching Models 10. A Unified Federated Framework for Trajectory Data Preparation via LLMs 11. TRIDENT: Cross-Domain Trajectory Spatio-Temporal Representation via Distance-Preserving Triplet Learning 12. CoLLMLight: Cooperative Large Language Model Agents for Network-Wide Traffic Signal Control 13. DecompGAIL: Learning Realistic Traffic Behaviors with Decomposed Multi-Agent Generative Adversarial Imitation Learning 14. Advancing Multi-agent Traffic Simulation via R1-Style Reinforcement Fine-Tuning 15. Plan-R1: Safe and Feasible Trajectory Planning as Language Modeling 16. BridgeDrive: Diffusion Bridge Policy for Closed-Loop Trajectory Planning in Autonomous Driving 17. OccDriver: Future Occupancy Guided Dual-branch Trajectory Planner in Autonomous Driving 18. Learning Dynamics Feature Representation via Policy Attention for Dynamic Path Planning in Urban Road Networks 19. UrbanVerse: Scaling Urban Simulation by Watching City-Tour Videos 20. UrbanFeel：A Comprehensive Benchmark for Temporal and Perceptual Understanding of City Scenes through Human Perspective 21. CityLens: Evaluating Large Vision-Language Models for Urban Socioeconomic Sensing 22. Urban Socio-Semantic Segmentation with Vision-Language Reasoning 23. CitySeeker: How Do VLMs Explore Embodied Urban Navigation with Implicit Human Needs?

1 Micro-Macro Coupled Koopman Modeling on Graph for Traffic Flow Prediction

链接 ：++https://openreview.net/forum?id=fhDqFk4DgI++

关键词：Koopman Operator; Traffic Flow Prediction

作者：Bairan Xiang, Chenguang Zhao, Huan Yu

分数：6, 6, 6, 6

信心：2, 2, 4, 3

均分 ：6.0

2 MoRA: Mobility as the Backbone for Geospatial Representation Learning at Scale

链接 ：++https://openreview.net/forum?id=IlBr5JJsCj++

关键词：GeoAI, spatial representation learning, location embedding, multi-modal, contrastive learning

作者：Ya Wen, Jixuan Cai, Qiyao Ma, Linyan Li, Xinhuan Chen, Chris Webster, Yulun Zhou

分数：8, 6, 6, 4

信心：4, 3, 4, 5

均分 ：6.0

TL; DR：We present MoRA, a human-centric geospatial framework that leverages a mobility graph as its core backbone to fuse various data modalities, aiming to learn embeddings that represent the socio-economic context and functional role of a location.

3 ELLMob: Event-Driven Human Mobility Generation with Self-Aligned LLM Framework

链接 ：++https://openreview.net/forum?id=MPYsaBgZIT++

关键词：Human Mobility Generation, Large Language Models, Event-Driven Mobility, Urban Computing

作者：Yusong Wang, Chuang Yang, Jiawei Wang, Xiaohang Xu, Jiayi Xu, Dongyuan Li, Chuan Xiao, Renhe Jiang

分数：6, 6, 4, 4

信心：2, 4, 4, 3

均分 ：5.0

4 Lightweight Spatio-Temporal Modeling via Temporally Shifted Distillation for Real-Time Accident Anticipation

链接 ：++https://openreview.net/forum?id=8zzfTSVds2++

关键词：lightweight spatio-temporal modeling, model distillation, accident anticipation, edge deployment

作者：Patrik Patera, Yie-Tarng Chen, Wen-Hsien Fang

分数：6, 6, 4

信心：2, 2, 2

均分 ：5.333333333333333

TL; DR：A lightweight, real-time accident predictor trained via novel temporally shifted distillation, combining efficient spatial encoding and recurrent temporal modeling, running on edge devices.

5 ST-HHOL: Spatio-Temporal Hierarchical Hypergraph Online Learning for Crime Prediction

链接 ：++https://openreview.net/forum?id=Nc3dl43s5Z++

关键词：Crime prediction, Spatio-temporal graph neural networks, Spatio-temporal data mining

作者：Keqing Du, Yufan Kang, Xinyu Yang, Wei Shao

分数：4, 6, 2, 8, 4

信心：3, 4, 4, 4, 4

均分 ：4.8

TL; DR：We propose ST-HHOL, an online spatio-temporal crime prediction framework that leverages hierarchical hypergraphs to uncover dual-specific patterns and tackle concept drift in non-stationary crime data.

6 A General Spatio-Temporal Backbone with Scalable Contextual Pattern Bank for Urban Continual Forecasting

链接 ：++https://openreview.net/forum?id=LHSea6DI8U++

关键词：general backbone, contextual pattern bank, continual spatio-temporal forecasting

作者：Aoyu Liu, Yaying Zhang

分数：4, 8, 6

信心：3, 5, 4

均分 ：6.0

7 USTBench: Benchmarking and Dissecting Spatiotemporal Reasoning Capabilities of LLMs as Urban Agents

链接 ：++https://openreview.net/forum?id=ETzBStUFJy++

关键词：large language model, spatiotemporal reasoning, urban science

作者：Siqi Lai, Yansong Ning, Zirui Yuan, Zhixi Chen, Hao Liu

分数：6, 6, 4, 8

信心：4, 3, 4, 4

均分 ：6.0

TL; DR：A benchmark for evaluating the urban spatiotemporal reasoning abilities of LLMs.

8 UrbanGraph: Physics-Informed Spatio-Temporal Dynamic Heterogeneous Graphs for Urban Microclimate Prediction

链接 ：++https://openreview.net/forum?id=ckjNF94cIi++

关键词：Spatio-Temporal Graph, Heterogeneous Graph, Dynamic Graph, Physics-Informed ML, Urban Microclimate

作者：Weilin Xin, Chenyu Huang, Peilin Li, Jing Zhong, Jiawei Yao

分数：4, 6, 6, 6

信心：4, 3, 3, 4

均分 ：5.5

9 TrajFlow: Nation-wide Pseudo GPS Trajectory Generation with Flow Matching Models

链接 ：++https://openreview.net/forum?id=BDOldEjwCE++

关键词：Flow matching, Human Trajectory, Generative modeling, Human mobility

作者：Peiran Li, Jiawei Wang, Haoran Zhang, Xiaodan Shi, Noboru Koshizuka, Chihiro Shimizu, Renhe Jiang

分数：10, 4, 6, 6

信心：5, 3, 5, 4

均分 ：6.5

TL; DR：This paper proposed TrajFM, a flow-matching-based GPS trajectory generation model that overcomes scale, diversity, and efficiency limitations of diffusion approaches to enable nationwide, multi-scale, and multi-modal human mobility data generation.

10 A Unified Federated Framework for Trajectory Data Preparation via LLMs

链接 ：++https://openreview.net/forum?id=MIelckWrEK++

关键词：Trajectory Data Preparation, Federated Learning, Large Language Model, Trajectory Preprocessing

作者：Zhihao Zeng, Ziquan Fang, Wei Shao, Lu Chen, Yunjun Gao

分数：6, 4, 8, 4

信心：3, 3, 5, 5

均分 ：5.5

11 TRIDENT: Cross-Domain Trajectory Spatio-Temporal Representation via Distance-Preserving Triplet Learning

链接 ：++https://openreview.net/forum?id=gOk3o4lMRD++

关键词：Spatiotemporal representation learning, Trajectory analysis, Cross-domain generalization, Triplet loss, Distance metric learning, self-supervised representation learning

作者：Guan Yi Jhang, Jeng-Chung Lien, Yu Hui-Ching, Hsu-Chao Lai, Jiun-Long Huang

分数：2, 6, 4, 6

信心：4, 3, 5, 4

均分 ：4.5

TL; DR：We learn self-supervised trajectory embedding with local pooling by fusing spatio-temporal features, and train with distance-preserving triplet loss aligning native-space 𝑑(𝑎,𝑝) and 𝑑(𝑎,𝑛), reduce distortion and improve cross-domain retrieval.

12 CoLLMLight: Cooperative Large Language Model Agents for Network-Wide Traffic Signal Control

链接 ：++https://openreview.net/forum?id=KeJqoEVOeY++

关键词：Traffic Signal Control, Large Language Model, Multi-Agent System, Intelligent Transportation

作者：Zirui Yuan, Siqi Lai, Hao Liu

分数：6, 4, 8, 4

信心：4, 4, 3, 3

均分 ：5.5

TL; DR：We introduce CoLLMLight, a cooperative LLM framework that achieves effective and efficient network-wide traffic signal control via spatiotemporal reasoning, asynchronous decision architecture, and cost-aware cooperation optimization.

13 DecompGAIL: Learning Realistic Traffic Behaviors with Decomposed Multi-Agent Generative Adversarial Imitation Learning

链接 ：++https://openreview.net/forum?id=AcDx2tUZPb++

关键词：traffic simulation, multi-agent imitation learning, generative adversarial imitation learning

作者：Ke Guo, Haochen Liu, XIAOJUN WU, Chen Lv

分数：4, 4, 8

信心：4, 5, 4

均分 ：5.333333333333333

14 Advancing Multi-agent Traffic Simulation via R1-Style Reinforcement Fine-Tuning

链接 ：++https://openreview.net/forum?id=7BiQwV9Sic++

关键词：Autonomous Driving, Reinforcement Fine-Tuning, Multi-agent Traffic Simulation

作者：Muleilan Pei, Shaoshuai Shi, Shaojie Shen

分数：8, 2, 6, 4

信心：3, 4, 4, 5

均分 ：5.0

TL; DR：A novel R1-style Reinforcement Fine-Tuning (RFT) paradigm for multi-agent traffic simulation in autonomous driving.

15 Plan-R1: Safe and Feasible Trajectory Planning as Language Modeling

链接 ：++https://openreview.net/forum?id=uusTA1rBhR++

关键词：Trajectory Planning, Reinforcement Learning, Autonomous Driving

作者：Xiaolong Tang, Meina Kan, Shiguang Shan, Xilin Chen

分数：4, 8, 8, 6

信心：2, 4, 4, 3

均分 ：6.5

TL; DR：We propose Plan-R1, a two-stage framework that decouples planning principle alignment from behavior learning to overcome the limitations of expert data. With VD-GRPO to preserve safety-critical signals, Plan-R1 achieves SOTA results on nuPlan.

16 BridgeDrive: Diffusion Bridge Policy for Closed-Loop Trajectory Planning in Autonomous Driving

链接 ：++https://openreview.net/forum?id=dJKhjK4zpp++

关键词：Diffusion policy, closed-loop planning, end-to-end autonomous driving

作者：Shu Liu, Wenlin Chen, Weihao Li, Zheng Wang, Lijin Yang, Jianing Huang, YipinZhang, Zhongzhan Huang, Ze Cheng, Hao Yang

分数：6, 6, 4, 6

信心：3, 5, 4, 4

均分 ：5.5

17 OccDriver: Future Occupancy Guided Dual-branch Trajectory Planner in Autonomous Driving

链接 ：++https://openreview.net/forum?id=abJCjkIwi5++

关键词：Autonomous Driving, Trajectory Planning

作者：Zhao Huang, Bowen Zhang, Zhongzhu Li, Di Lin

分数：2, 6, 8

信心：4, 3, 5

均分 ：5.333333333333333

18 Learning Dynamics Feature Representation via Policy Attention for Dynamic Path Planning in Urban Road Networks

链接 ：++https://openreview.net/forum?id=1E4Bltg6Xb++

关键词：Dynamic Path Planning; Reinforcement Learning; State Representation; Dynamics Feature Representation; Policy Attention Mechanism

作者：Kai Zhang, Jingjing Gu, Qiuhong Wang

分数：2, 6, 6

信心：5, 3, 3

均分 ：4.666666666666667

19 UrbanVerse: Scaling Urban Simulation by Watching City-Tour Videos

链接 ：++https://openreview.net/forum?id=HE6j2jtjII++

关键词：Simulation, Real-to-Sim, Sim-to-Real, Digital Twin, Robot Navigation, Reinforcement Learning

作者：Mingxuan Liu, Honglin He, Elisa Ricci, Wayne Wu, Bolei Zhou

分数：6, 6, 2, 8

信心：3, 5, 5, 4

均分 ：5.5

20 UrbanFeel：A Comprehensive Benchmark for Temporal and Perceptual Understanding of City Scenes through Human Perspective

链接 ：++https://openreview.net/forum?id=OtLC2JNGZf++

关键词：Benchmark, Urban Change, Urban Perception, Multimodel Large Language Models

作者：Jun He, Yi Lin, Zilong Huang, Jiacong Yin, Junyan Ye, Yuchuan Zhou, Weijia Li, Xiang Zhang

分数：6, 4, 4, 6

信心：5, 4, 4, 4

均分 ：5.0

21 CityLens: Evaluating Large Vision-Language Models for Urban Socioeconomic Sensing

链接 ：++https://openreview.net/forum?id=kswX9NfAlo++

关键词：Multi-modal Large Language Model, Socioeconomic Prediction, Urban Imagery, Urban Science, Benchmark

作者：Tianhui Liu, Hetian Pang, Xin Zhang, Tianjian Ouyang, Zhiyuan Zhang, Jie Feng, Yong Li, Pan Hui

分数：6, 6, 6, 2

信心：4, 4, 3, 4

均分 ：5.0

TL; DR：We propose a global scale benchmark to evaluate the performance of large language-vision models for urban imagery-based socioeconomic prediction

22 Urban Socio-Semantic Segmentation with Vision-Language Reasoning

链接 ：++https://openreview.net/forum?id=sVN9K0BLQj++

关键词：Remote Sensing, Semantic Segmentation, Vision Language Model, Reinforcement Learning

作者：Yu Wang, Yi Wang, Rui Dai, Yujie Wang, Kaikui Liu, Xiangxiang Chu, Yansheng Li

分数：6, 2, 2, 4, 6

信心：3, 3, 4, 4, 4

均分 ：4.0

23 CitySeeker: How Do VLMs Explore Embodied Urban Navigation with Implicit Human Needs?

链接 ：++https://openreview.net/forum?id=hzf23XSDcs++

关键词：Embodied Urban Navigation, Vision-Language Models, Urban Intelligence, Spatial Cognition

作者：Siqi Wang, Chao Liang, Yunfan Gao, Erxin Yu, Sen Li, Jing Li, Haofen Wang

分数：6, 4, 2, 4

信心：5, 4, 4, 4

均分 ：4.0