Development of a decomposition-optimization-transformer hybrid model for spatiotemporal forecasting of PM2.5 air...

The existing pollutants in the air, particularly fine particulate matter, coarse particulate matter, nitrogen dioxide, carbon monoxide, and ozone, pose remarkable health risks and environmental problems, particularly in rapidly urbanizing regions. Precise prediction of these pollutants is a vital step for impactful air quality management and proactive public health measures. To overcome these obstacles, the presented investigations bring out a novel Multivariate Empirical Mode Decomposition-Lévy Shrinkage-assisted Adaptive Differential Evolution-Transformer framework for predicting multiple pollutants. The model merges multivariate empirical mode decomposition with Lévy Shrinkage-based adaptive differential evolution for hyperparameter optimization, which permits effective modelling of the intricate interactions among pollutants and meteorological variables. This framework demonstrates reliable predictive performance across four major Chinese regions, such as Beijing, Guangzhou, Shanghai, and Shenzhen. For instance, in Beijing, the model achieves a coefficient of determination of 0.98 for PM2.5 prediction, with a root mean square error of 3.04 µg/m³ and a mean absolute error of 1.96 µg/m³ . The model also exhibits strong adaptability in other cities, with the adopted coefficients of determination of 0.97-0.98. Additionally, the model integrates spatial-temporal analysis using data from 12 monitoring stations in Beijing, capturing both temporal fluctuations and spatial heterogeneity. Moreover, it predicts the Air Quality Index (AQI) based on PM2.5 levels, classifying air quality like Good to Hazardous, and provides insights into health risks for vulnerable populations. The model's multi-pollutant capabilities enable comprehensive health risk assessments, which could be linked to epidemiological studies. Using walk-forward cross-validation, the model ensures robust generalization and adaptability. The results highlight the model's effectiveness in predicting air quality, showing its potential to be a valuable tool for urban air quality management and public health protection.