.png){kind=icon}
.png){kind=icon}
Return
Predicting Fine Spatial-Temporal Scale Bus Emissions Using Graph Embedding Deep Learning Model
.png){kind=icon}
PRE
AI
DOI：10.3808/jei.202500534.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Emissions from urban transportation significantly impact the environment and public health, including air pollution, increased carbon emissions, and various health issues. Public transportation is vital in alleviating traffic congestion and reducing overall emissions. However, due to the complex urban environment, bus emissions exhibit spatial heterogeneity at fine scales, and the influencing variables have nonlinearity and complexity, making it difficult to predict real-time emissions accurately. To address this challenge, we used real-time emission data from hybrid electric vehicle buses (HEBs), which are widely used in China, to train four models: linear regression (LR), random forest (RF), XGBOOST, and graph embedding-based generative adversarial network (GE-GAN). These models predicted real-time emissions at fine scales and analyzed the interactions between emission variables. Using graph embedding and GAN algorithms, the GE-GAN model effectively captures semantic and spatial information between variables, understanding complex interactions. Among all the models, GE-GAN achieved the best results in validation tests. For CO2 prediction, the root mean square error (RMSE) was 0.98, and the mean absolute error (MAE) was 1.31. For CO prediction, the RMSE was 0.03, and the MAE was 0.08. For NOx prediction, the RMSE was 0.43, and the MAE was 0.57. The experimental results demonstrated that GE-GAN can accurately predict real-time bus emissions at a fine scale in mixed urban areas. The accurate predictions can assist planning departments in determining pollutant distribution in urban areas, optimizing public transportation routes, and contributing to reducing city carbon emissions.
Keywords:
bus emission
graph embedding
generative adversarial networks
real-time prediction
Journal
J
IF:
5.4
Papers:
529
Citations:
661
View more organizations