A point–path collaborative imputation method for missing trajectories of long-distance freight

Trajectory data from long-distance freight vehicles often contain missing segments due to blockages, signal blind zones, and equipment failures, which hinder real-time supervision and increase operational risks. These trajectories exhibit strong regularity from fixed routes and stable path preferences, enabling point-path collaborative imputation. This paper proposes a Point-Path Collaborative Imputation Method (PPCIM) integrating a Heterogeneous Graph Attention Network (HAN) and a Conditional Generative Adversarial Network (CGAN), reformulating trajectory reconstruction as a joint point-path optimization problem. Soft-DTW first identifies similar historical trajectories to generate initial imputation points; HAN then estimates path selection probabilities and node importance to capture spatiotemporal dependencies; a CGAN with a dual-head discriminator refines the reconstruction. Using Liquefied Petroleum Gas transport data from Foshan, China, PPCIM reduces MAE and RMSE by 80.2% and 81.3% over traditional two-stage methods, with path recognition precision and recall improved by 73.9% and 61.0%. Tests across 29 cities confirm strong robustness and accuracy.