[Paper Review] Dynamic Planning of Bicycle Stations in Dockless Public Bicycle-sharing System Using Gated Graph Neural Network

Jianguo Chen, Kenli Li+3 more|arXiv (Cornell University)|Jan 19, 2021

Urban Transport and Accessibility33 references4 citations

TL;DR

This paper proposes a Dynamic Planning of Bicycle Stations (BSDP) system for dockless public bicycle-sharing systems using a Gated Graph Neural Network (GGNN) to predict optimal station layouts. By clustering drop-off locations, modeling them as time-evolving weighted digraphs, and training on graph sequence data, the system achieves high prediction accuracy (AUC of 0.91) and enables dynamic, revenue-optimized, and city-management-compliant station planning.

ABSTRACT

Benefiting from convenient cycling and flexible parking locations, the Dockless Public Bicycle-sharing (DL-PBS) network becomes increasingly popular in many countries. However, redundant and low-utility stations waste public urban space and maintenance costs of DL-PBS vendors. In this paper, we propose a Bicycle Station Dynamic Planning (BSDP) system to dynamically provide the optimal bicycle station layout for the DL-PBS network. The BSDP system contains four modules: bicycle drop-off location clustering, bicycle-station graph modeling, bicycle-station location prediction, and bicycle-station layout recommendation. In the bicycle drop-off location clustering module, candidate bicycle stations are clustered from each spatio-temporal subset of the large-scale cycling trajectory records. In the bicycle-station graph modeling module, a weighted digraph model is built based on the clustering results and inferior stations with low station revenue and utility are filtered. Then, graph models across time periods are combined to create a graph sequence model. In the bicycle-station location prediction module, the GGNN model is used to train the graph sequence data and dynamically predict bicycle stations in the next period. In the bicycle-station layout recommendation module, the predicted bicycle stations are fine-tuned according to the government urban management plan, which ensures that the recommended station layout is conducive to city management, vendor revenue, and user convenience. Experiments on actual DL-PBS networks verify the effectiveness, accuracy and feasibility of the proposed BSDP system.

Motivation & Objective

• Address the problem of inefficient, redundant, and poorly managed bicycle station deployments in dockless public bicycle-sharing (DL-PBS) systems.
• Overcome limitations of static, experience-based station planning in traditional systems by leveraging real-time, large-scale trajectory data.
• Develop a dynamic, data-driven system that adapts to changing urban demand patterns and supports both vendor profitability and urban management goals.
• Enable accurate prediction of future bicycle station locations and usage to guide optimal deployment and dispatching strategies.

Proposed method

• Cluster large-scale bicycle drop-off locations using the DADC algorithm to identify candidate station sites from spatio-temporal trajectory data.
• Construct a weighted directed graph model based on clustering results, filtering out low-revenue and low-utility stations to improve model relevance.
• Integrate graph models across time periods into a graph sequence model to capture temporal evolution of station usage patterns.
• Train a Gated Graph Neural Network (GGNN) on the graph sequence data to learn temporal dependencies and predict future station locations and demand.
• Fine-tune predicted station layouts using government urban management plans to ensure feasibility, legality, and alignment with city planning.
• Use the final recommended layout for downstream applications such as bicycle dispatching, trajectory tracking, and fault detection.

Experimental results

Research questions

• RQ1 How can large-scale, spatio-temporal bicycle trajectory data be effectively transformed into actionable candidate station locations?
• RQ2 What graph-based modeling approach best captures the dynamic evolution of bicycle station usage across time in a dockless system?
• RQ3 To what extent can a Gated Graph Neural Network (GGNN) outperform other models (e.g., GCN, GNN, LSTM) in predicting future bicycle station locations?
• RQ4 How can predicted station layouts be adjusted to align with urban management policies while maximizing service quality and vendor revenue?

Key findings

• The DADC clustering method outperformed DPC and DBSCAN, achieving an average AUC of 0.82, with performance improving significantly as data volume increased.
• The GGNN model achieved the highest prediction accuracy among all compared models, with an average AUC of 0.91 after 100 training iterations, surpassing GNN (0.83), GCN (0.80), and LSTM (0.71).
• GGNN demonstrated the fastest convergence and lowest error, achieving a minimum average RMSE of 0.13 after 100 iterations, significantly lower than GNN (0.26), GCN (0.29), and LSTM (0.33).
• The BSDP system successfully generated feasible, revenue-optimized, and city-compliant station layouts, demonstrating its practical feasibility and integration potential.
• The system’s ability to process real-world DL-PBS datasets confirms its scalability and robustness across different urban administrative regions.
• The integration of government urban planning constraints into the recommendation phase ensures that proposed layouts are both operationally viable and policy-compliant.