Core Concepts
Integrating multi-source transportation demand data with a spatial-temporal Transformer model can significantly improve the accuracy of urban parking availability prediction.
Abstract
This study proposes a parking availability prediction framework that integrates spatial-temporal deep learning with multi-source data fusion, including traffic demand data from metro, bus, taxi, and online ride-hailing services, as well as parking lot data.
The key highlights are:
A clustering-based method is used to establish "parking cluster zones" that group similar parking lots within certain urban areas. This allows the model to capture the spatial-temporal parking dynamics and correlations among related parking lots.
The multi-source transportation demand data (metro, bus, taxi, online ride-hailing) within the parking cluster zones is integrated as important spatial-temporal features for the parking availability prediction.
A Transformer-based spatial-temporal deep learning model is developed and customized for the parking prediction task. It outperforms various baseline models including traditional statistical, machine learning, and other deep learning approaches in terms of prediction accuracy metrics like MSE, MAE, and MAPE.
The ablation study demonstrates the effectiveness of the integrated multi-source demand features and the advantages of the Transformer model in capturing long-term spatial-temporal dependencies for both short-term and long-term parking availability forecasting.
Overall, the proposed framework leverages the power of multi-source data fusion and advanced deep learning techniques to deliver more accurate and timely parking availability predictions, which can benefit both drivers and urban planners for efficient and sustainable urban mobility.
Stats
The parking availability prediction models were evaluated using the following key metrics:
Mean Squared Error (MSE)
Mean Absolute Error (MAE)
Mean Absolute Percentage Error (MAPE)
The Transformer model achieved the best performance with:
MSE: 0.0626
MAE: 0.1358
MAPE: 0.5496
Quotes
"By fusing multi-source demanding data with spatial-temporal deep learning techniques, this approach offers the potential to develop parking availability prediction systems that furnish more accurate and timely information to both drivers and urban planners, thereby fostering more efficient and sustainable urban mobility."