Self-Supervised Spatiotemporal Clustering of Vehicle Emissions With Graph Convolutional Network

Area of Science:

• Environmental Science
• Computer Science
• Data Science

Background:

• Spatiotemporal clustering of vehicle emissions is crucial for understanding road traffic air pollution evolution.
• Existing Graph Convolutional Network (GCN) methods inadequately model the interplay between spatial and temporal emission variations.
• This limitation leads to incomplete descriptions and inaccurate detection of air pollution patterns.

Purpose of the Study:

• To propose a novel two-way self-supervised spatiotemporal representation learning scheme for vehicle emissions.
• To address the challenge of unsupervised learning in accurately detecting air pollution evolution patterns.
• To improve the accuracy of spatiotemporal clustering by capturing the interactions between spatial and temporal features.

Main Methods:

• A two-way self-supervised scheme progressively learns temporal and spatial features in a mutually reinforced manner.
• Initial temporal representations are captured using a pretrained BiLSTM network.
• A Graph Convolutional Network (GCN) is employed for feature clustering, constrained by a two-way self-supervised mechanism and refined through joint optimization.

Main Results:

• The proposed method demonstrates superior performance compared to state-of-the-art approaches in spatiotemporal clustering of vehicle emissions.
• Experimental results on the Xian city traffic emission dataset (2020) validate the effectiveness of the approach.
• The method successfully captures and refines the evolution patterns of air pollution from road traffic.

Conclusions:

• The developed two-way self-supervised learning scheme effectively models spatiotemporal correlations in vehicle emissions.
• This approach overcomes the limitations of existing methods by integrating spatial and temporal feature interactions.
• The findings offer a significant advancement in accurately detecting and analyzing air pollution evolution patterns.