Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AirPhyNet: Harnessing Physics-Guided Neural Networks for Air Quality Prediction

Kethmi Hirushini Hettige, Jiahao Ji, Shili Xiang, Cheng Long, Gao Cong, Jingyuan Wang

TL;DR

The paper tackles accurate, interpretable PM2.5 forecasting in urban air quality, especially under data sparsity. It proposes AirPhyNet, a physics-guided neural network that embeds diffusion and advection through a Diffusion-Advection differential equation network within a graph-based, end-to-end learning framework. On Beijing and Shenzhen datasets, AirPhyNet outperforms state-of-the-art baselines across 24h–72h forecasts and in sudden-change scenarios, with improvements up to around 10% in predictive error. Ablation and case-study analyses demonstrate the necessity of combining diffusion and advection and show the model yields physically meaningful predictions aligned with wind fields.

Abstract

Air quality prediction and modelling plays a pivotal role in public health and environment management, for individuals and authorities to make informed decisions. Although traditional data-driven models have shown promise in this domain, their long-term prediction accuracy can be limited, especially in scenarios with sparse or incomplete data and they often rely on black-box deep learning structures that lack solid physical foundation leading to reduced transparency and interpretability in predictions. To address these limitations, this paper presents a novel approach named Physics guided Neural Network for Air Quality Prediction (AirPhyNet). Specifically, we leverage two well-established physics principles of air particle movement (diffusion and advection) by representing them as differential equation networks. Then, we utilize a graph structure to integrate physics knowledge into a neural network architecture and exploit latent representations to capture spatio-temporal relationships within the air quality data. Experiments on two real-world benchmark datasets demonstrate that AirPhyNet outperforms state-of-the-art models for different testing scenarios including different lead time (24h, 48h, 72h), sparse data and sudden change prediction, achieving reduction in prediction errors up to 10%. Moreover, a case study further validates that our model captures underlying physical processes of particle movement and generates accurate predictions with real physical meaning.

AirPhyNet: Harnessing Physics-Guided Neural Networks for Air Quality Prediction

TL;DR

The paper tackles accurate, interpretable PM2.5 forecasting in urban air quality, especially under data sparsity. It proposes AirPhyNet, a physics-guided neural network that embeds diffusion and advection through a Diffusion-Advection differential equation network within a graph-based, end-to-end learning framework. On Beijing and Shenzhen datasets, AirPhyNet outperforms state-of-the-art baselines across 24h–72h forecasts and in sudden-change scenarios, with improvements up to around 10% in predictive error. Ablation and case-study analyses demonstrate the necessity of combining diffusion and advection and show the model yields physically meaningful predictions aligned with wind fields.

Abstract

Air quality prediction and modelling plays a pivotal role in public health and environment management, for individuals and authorities to make informed decisions. Although traditional data-driven models have shown promise in this domain, their long-term prediction accuracy can be limited, especially in scenarios with sparse or incomplete data and they often rely on black-box deep learning structures that lack solid physical foundation leading to reduced transparency and interpretability in predictions. To address these limitations, this paper presents a novel approach named Physics guided Neural Network for Air Quality Prediction (AirPhyNet). Specifically, we leverage two well-established physics principles of air particle movement (diffusion and advection) by representing them as differential equation networks. Then, we utilize a graph structure to integrate physics knowledge into a neural network architecture and exploit latent representations to capture spatio-temporal relationships within the air quality data. Experiments on two real-world benchmark datasets demonstrate that AirPhyNet outperforms state-of-the-art models for different testing scenarios including different lead time (24h, 48h, 72h), sparse data and sudden change prediction, achieving reduction in prediction errors up to 10%. Moreover, a case study further validates that our model captures underlying physical processes of particle movement and generates accurate predictions with real physical meaning.
Paper Structure (22 sections, 28 equations, 4 figures, 4 tables)

This paper contains 22 sections, 28 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Air Quality Prediction Problem
  5. Physical dynamics of Air Pollutant Transport
  6. Diffusion-Advection Differential Equation
  7. Physics Guided Differential Equation Network
  8. Experiments
  9. Data Description
  10. Experimental Settings
  11. Performance Comparison
  12. Ablation Study
  13. Case Study
  14. Conclusion and Future Work
  15. Appendix
  16. ...and 7 more sections

Figures (4)

  • Figure 1: The overall architecture of AirPhyNet framework consisting of a RNN Encoder, GNN based DE Network and a Decoder. Heatmap indicate the PM2.5 concentrations while the dashed arrows represent the air pollutant transport between nodes due to diffusion and advection.
  • Figure 2: Results of Ablation Study
  • Figure 3: Visualization of predicted PM2.5 concentrations and wind direction. Heatmap represents the PM2.5 concentrations and the arrows indicate the wind direction
  • Figure 4: Visualization of predicted PM2.5 concentrations and diffusion from Station_3 on 2017-11-13 14:00.The line segments represent diffusion from Station_3 to its neighbouring stations and the magnitude of diffusion is reflected by the line colour.