Abstract: Air pollution affects human health and environmental ecosystems, and forecasting air quality is a crucial part of in-depth air pollution prevention and control, which is particularly vital for enhancing the defense capability against heavy polluted weather and protecting puplic health and safety. Conventional air quality forecasting methods are usually based on numerical or statistical models, however, these conventional forecasting methods are often limited by the complexity of models and incompleteness of data. With the gradual maturation of artificial intelligence technology, deep learning has brought new ideas for air quality forecasting research. This paper focuses on the current development status of deep learning-based models and introduces the effectiveness and potential of deep learning methods in air quality forecasting from four perspectives: time-dependent modeling, spatiotemporal correlation modeling, geographic information modeling, and hybrid data-driven and physics-based modeling frameworks. In addition, the auxiliary features and public datasets that support deep learning research are also listed. Finally, a summary outlook on further exploration and optimization of technical methods of deep learning models and their application prospects in air quality forecasting is presented.

Key words: air pollution monitoring, air quality prediction, deep learning, modeling