Multi-view multi-task spatiotemporal graph convolutional network for air quality prediction.

Area of Science:

• Environmental Science
• Computer Science
• Data Science

Background:

• Accurate air quality prediction is vital for intelligent cities, aiding environmental governance and public health.
• Existing models often struggle with complex intra-sensor and inter-sensor correlations, limiting prediction accuracy.
• Previous approaches primarily focused on spatial and temporal correlations, overlooking other relational aspects.

Purpose of the Study:

• To develop an advanced model for accurate air quality prediction in intelligent cities.
• To address the limitations of existing methods by incorporating multi-view and multi-task learning.
• To improve the understanding and modeling of complex correlations in air quality data.

Main Methods:

• Proposed a multi-view, multi-task spatiotemporal graph convolutional network (M2).
• Encoded three views: spatial (GCN for geographic correlation), logical (GCN for semantic correlation), and temporal (GRU for historical data correlation).
• Employed a multi-task learning paradigm with joint classification (air quality level) and regression (air quality value) tasks.

Main Results:

• The M2 model demonstrated superior performance compared to state-of-the-art methods on two real-world air quality datasets.
• The multi-view approach effectively captured diverse correlations, enhancing prediction accuracy.
• The multi-task learning paradigm improved the joint prediction of air quality levels and values.

Conclusions:

• The proposed M2 model offers a significant advancement in air quality prediction technology.
• Integrating spatial, logical, and temporal views, along with multi-task learning, is effective for complex environmental forecasting.
• This approach holds promise for enhancing environmental monitoring and decision-making in intelligent urban environments.