An Approach to Mine and Visual Analytics the Propagation Patterns of Traffic Congestion based on Graph Representation Learning

Traffic congestion seriously affects the transportation of residents and the sustainable development of society. How to mine traffic congestion propagation patterns from massive traffic data and formulate a scientific traffic planning scheme is a critical issue. Therefore, a visual analytics method for traffic congestion propagation pattern is proposed. Firstly, congestion events are defined and detected based on the speed reduction index, and spatial and temporal nearest neighbor constraints are used to identify the congestion propagation. Then, an algorithm is proposed to characterize congestion events by integrating node2vec and VGAE models, combined with dimensionality reduction and clustering algorithms, to explore congestion propagation patterns. Finally, a congestion propagation patterns visual analytics system with multi-view linkage is designed to support users to perceive the congestion density and explore the congestion propagation patterns interactively. Based on 14 consecutive days of GPS trajectory data in a city, a comparative experiment using the average accuracy metric is conducted, and the result shows that the proposed method has higher accuracy in reconstructing congestion propagation. Three experts in the fields of transport and visualization are invited to evaluate the proposed system, and subjective scores are given to the ease of use, effectiveness and usefulness of the system through two case studies, including the task of global congestion sensing and the task of analyzing the congestion propagation patterns on specific roads. Quantitative evaluation result shows that the system can effectively assist users in analyzing congestion propagation patterns and provide a basis for formulating traffic management schemes.