Spatio-temporal Parking Behaviour Forecasting and Analysis Before and During COVID-19

Shuhui Gong, Xiaopeng Mo, Rui Cao, Yu Liu, Wei Tu, Ruibin BaiarXiv 2021 · arXiv

Abstract

This paper introduces a spatial-aware framework for predicting urban parking occupancy by modeling the hidden correlations between nearby parking lots using a Temporal Graph Convolutional Network (T-GCN). The researchers demonstrate that incorporating spatial dependencies significantly improves forecasting accuracy, particularly during the temporal irregularities caused by the COVID-19 pandemic.

Introduction

Predicting parking demand is vital for reducing urban traffic congestion and improving city planning. However, most existing models treat parking lots as isolated entities, focusing solely on historical time-series data. This approach overlooks two critical factors: spatial dependencies and unexpected disruptions.

Unlike road networks, parking lots lack direct physical connections, making their spatial correlations harder to quantify. Yet, human behavior suggests that if one lot is full, a driver will likely choose another within a tolerable walking distance (typically ~500 meters). Furthermore, purely temporal models often fail when routines are shattered, such as during the COVID-19 pandemic. This research addresses these gaps by proposing a model that understands how different locations influence one another.

Methodology: The T-GCN Framework

The authors propose a two-step spatial-aware framework to capture both the geography and the history of parking occupancy.

First, a Connection Graph is constructed using a weighted adjacency matrix. Instead of a binary connection, the model uses a distance decay function. If two lots are within a threshold ($d_0 = 500m$), they are assigned a connection strength that weakens as the distance between them increases.

Second, the Temporal Graph Convolutional Network (T-GCN) processes this data. The GCN layer captures spatial features by aggregating information from neighboring parking lots in the graph, while the Gated Recurrent Unit (GRU) layer processes the GCN's output to capture the temporal evolution of occupancy.

The proposed T-GCN framework integrates a distance decay function to model spatial correlations between parking lots with a GRU for temporal forecasting.

Data Analysis and Spatial Insights

The study utilizes a real-world dataset from Ningbo, China, containing over one million records from 136 parking lots. The analysis revealed a "Noon Peak" phenomenon in pre-pandemic data, where occupancy peaked sharply between 11 am and 12 pm as drivers preferred to park during the lunch hour to avoid peak traffic.

Spatial distribution of parking occupancy in Ningbo, highlighting high-demand hotspots during a typical weekday.

During the COVID-19 pandemic, parking hotspots shifted significantly from business districts and entertainment centers to residential communities. Despite these shifts, experimental results confirmed that a 500-meter threshold for the connection graph yielded the best results, aligning with urban planning theories regarding tolerable walking distances.

Performance Evaluation

The proposed T-GCN framework was compared against several baselines, including ARIMA, SVR, ANN, and a standard GRU. The framework achieved the lowest errors (RMSE, MAE, MAPE) across all scenarios.

Notably, the improvement over the standard GRU—which only considers time—was much more pronounced during the COVID-19 period. For example, during the pandemic, the framework's RMSE was 0.09 for weekdays, compared to the GRU's 0.13. This suggests that when temporal patterns become irregular, the spatial component of the model acts as a vital stabilizer.

Conclusion

By modeling parking lots as a connected graph rather than isolated points, the T-GCN framework provides a more resilient and accurate forecasting tool. Future work will aim to incorporate actual walking paths and external variables like weather and real-time traffic flow to further refine the spatial weights and behavioral analysis.

Summary updated

Abstract

Parking demand forecasting and behaviour analysis have received increasing attention in recent years because of their critical role in mitigating traffic congestion and understanding travel behaviours. However, previous studies usually only consider temporal dependence but ignore the spatial correlations among parking lots for parking prediction. This is mainly due to the lack of direct physical connections or observable interactions between them. Thus, how to quantify the spatial correlation remains a significant challenge. To bridge the gap, in this study, we propose a spatial-aware parking prediction framework, which includes two steps, i.e. spatial connection graph construction and spatio-temporal forecasting. A case study in Ningbo, China is conducted using parking data of over one million records before and during COVID-19. The results show that the approach is superior on parking occupancy forecasting than baseline methods, especially for the cases with high temporal irregularity such as during COVID-19. Our work has revealed the impact of the pandemic on parking behaviour and also accentuated the importance of modelling spatial dependence in parking behaviour forecasting, which can benefit future studies on epidemiology and human travel behaviours.
BibTeX
@misc{gong2021spatio,
  author = {Shuhui Gong and Xiaopeng Mo and Rui Cao and Yu Liu and Wei Tu and Ruibin Bai},
  title = {{Spatio-temporal Parking Behaviour Forecasting and Analysis Before and During COVID-19}},
  year = {2021},
  eprint = {2108.07731},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2108.07731},
}
Figure