Predicting Station-level Hourly Demands in a Large-scale Bike-sharing Network: A Graph Convolutional Neural Network Approach (1712.04997v2)

Abstract: This study proposes a novel Graph Convolutional Neural Network with Data-driven Graph Filter (GCNN-DDGF) model that can learn hidden heterogeneous pairwise correlations between stations to predict station-level hourly demand in a large-scale bike-sharing network. Two architectures of the GCNN-DDGF model are explored; GCNNreg-DDGF is a regular GCNN-DDGF model which contains the convolution and feedforward blocks, and GCNNrec-DDGF additionally contains a recurrent block from the Long Short-term Memory neural network architecture to capture temporal dependencies in the bike-sharing demand series. Furthermore, four types of GCNN models are proposed whose adjacency matrices are based on various bike-sharing system data, including Spatial Distance matrix (SD), Demand matrix (DE), Average Trip Duration matrix (ATD), and Demand Correlation matrix (DC). These six types of GCNN models and seven other benchmark models are built and compared on a Citi Bike dataset from New York City which includes 272 stations and over 28 million transactions from 2013 to 2016. Results show that the GCNNrec-DDGF performs the best in terms of the Root Mean Square Error, the Mean Absolute Error and the coefficient of determination (R2), followed by the GCNNreg-DDGF. They outperform the other models. Through a more detailed graph network analysis based on the learned DDGF, insights are obtained on the black box of the GCNN-DDGF model. It is found to capture some information similar to details embedded in the SD, DE and DC matrices. More importantly, it also uncovers hidden heterogeneous pairwise correlations between stations that are not revealed by any of those matrices.

• The paper introduces a GCNN-DDGF model that captures hidden spatial-temporal correlations in bike-sharing networks.
• Integrating an LSTM block, the approach effectively models temporal dependencies, achieving an RMSE of 2.12 on NYC Citi Bike data.
• Graph network analysis reveals that the DDGF uncovers latent inter-station relationships not captured by conventional data matrices.

Predicting Station-level Hourly Demand in Bike-sharing Networks Using GCNN-DDGF

The paper presents an innovative approach to predict station-level hourly demand in large-scale bike-sharing networks through the application of a Graph Convolutional Neural Network with Data-Driven Graph Filter (GCNN-DDGF). This paper addresses the inherent challenges in capturing hidden correlations between stations, both spatially and temporally, through a deep learning model that adeptly learns from raw data without predefined adjacency matrices.

This research investigates two configurations of the GCNN-DDGF model: the regular GCNN-DDGF (GCNN-DDGF_reg) and the GCNN-DDGF with an embedded Long Short-term Memory (LSTM) recurrent block (GCNN-DDGF_rec). The LSTM integration specifically facilitates the consideration of temporal dependencies in demand series. Moreover, four distinct GCNN variants are introduced based on numerous bike-sharing system data matrices such as Spatial Distance (SD), Average Trip Duration (ATD), Demand Correlation (DC), and Demand (DE) matrices.

The empirical evaluation is conducted using data from the Citi Bike system in New York City, covering 272 stations and 28 million transactions from 2013 to 2016. This extensive dataset is processed to develop demand prediction models whose performances are evaluated on a testing set through key metrics like Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and R² coefficient of determination. Notably, the GCNN-DDGF_rec configuration exhibited superior performance, achieving a minimum RMSE of 2.12, outperforming conventional predictive models including LSTM, MLP, and XGBoost. This finding suggests considerable improvements in prediction accuracy when employing dynamic graph convolutional frameworks capable of accounting for both spatial and temporal dependencies.

The paper also explores graph network analysis leveraging the DDGF to unravel the "black box" element of the neural network's functioning. Analysis via Gephi indicates that the DDGF captures hidden station correlations invisible to the SD, DE, and DC matrices alone, thus underscoring the broader capability of the GCNN-DDGF model to abstract nuanced spatial-temporal demand patterns.

For researchers in artificial intelligence and transportation planning, this paper highlights a practical application of GCNNs in dynamic environments, enhancing operational efficiencies in bike-sharing systems—a critical component of sustainable urban transportation infrastructures. Methodologically, the GCNN-DDGF presents an enabling framework for future work seeking to apply generative graph-based learning in various transportation network prediction scenarios, potentially integrating additional external variables like weather patterns or event occurrence, thereby expanding the current model's predictive robustness.

Implications for AI-centered development suggest not only an expansion into other forms of shared mobility services but also the adaptation of the GCNN architecture for real-time adaptive systems that can dynamically update and tune model parameters. Future research directions may explore model extensibility to directed graph settings, offering more intricate modeling of asymmetric station relationships, and a focus on the interpretability and transparency of AI models that underpin city-scale transportation logistics.