Related Experiment Video
Updated: Jul 27, 2026
16:14
Trajectory Data Analyses for Pedestrian Space-time Activity Study
Published on: February 25, 2013
13.5K
Spatio-temporal data prediction of multiple air pollutants in multi-cities based on 4D digraph convolutional neural
Li Wang1, Qianhui Tang1, Xiaoyi Wang2
1Beijing Laboratory for Intelligent Environmental Protection, School of Artificial Intelligence, Beijing Technology and Business University, Beijing, China.
Plos One
|December 22, 2023
Summary
This study introduces a novel four-dimensional directed graph convolutional network-long short-term memory (4D-DGCN-LSTM) model for accurate multi-city, multi-pollutant air quality forecasting. The model significantly improves prediction accuracy by capturing complex spatio-temporal correlations.
Area of Science:
- Environmental Science and Engineering
- Artificial Intelligence
- Data Science
Background:
- Current one-dimensional undirected graph neural networks struggle to capture complex two-dimensional spatial and directed correlations in multi-city pollutant prediction.
- Accurate forecasting of atmospheric pollutant concentrations is crucial for public health and environmental management.
Purpose of the Study:
- To develop a novel four-dimensional directed graph convolutional network-long short-term memory (4D-DGCN-LSTM) model for enhanced multi-city, multi-pollutant prediction.
- To effectively capture two-dimensional spatial directed information, node correlations, and temporal dynamics in atmospheric pollutant data.
Main Methods:
- Construction of a four-dimensional directed graph convolutional network (4D-DGCN) model incorporating geographical city locations.
- Application of spectral decomposition and tensor operations for graph Fourier analysis.
- Integration of a long short-term memory (LSTM) network for spatio-temporal feature extraction and pollutant concentration prediction.
Main Results:
- The 4D-DGCN-LSTM model demonstrated superior performance in predicting atmospheric pollutant concentrations compared to existing models.
- Significant Mean Absolute Error (MAE) reductions of 1.12% (vs. 4D-DGCN), 4.91% (vs. GCN-LSTM), 5.62% (vs. GCN), and 11.67% (vs. LSTM) were achieved.
- Canonical correlation analysis confirmed the temporal, spatial, and multi-factor correlations within the Taihu Lake city cluster's 2020 atmospheric data.
Conclusions:
- The proposed 4D-DGCN-LSTM model effectively addresses the limitations of previous methods by integrating directed spatial and temporal information.
- This advanced model offers a robust solution for accurate and reliable prediction of multiple atmospheric pollutants across different cities.
- The findings highlight the potential of sophisticated graph neural networks and LSTMs in environmental big data analytics and forecasting.
