Attention scores and peak perception in long-term ozone prediction using deep learning

To address the limitations of traditional ozone (O 3 ) forecasting models, this study established a novel Transformer-based model integrating attention scores and the peak perception. Attention scores dynamically quantify nonlinear relationships between O 3 and influencing factors, while peak perception method penalizes peak O 3 errors, ensuring accurate predictions during O 3 exceedance events. Our results demonstrate that the proposed model significantly improves the prediction accuracy for both hourly and maximum daily 8-h average O 3 concentrations, with the R 2 value increasing from 0.56 to 0.83 and the mean squared error decreasing from 0.47 to 0.42. Wind direction, wind speed, and carbon monoxide emerged as dominant factors during pollution episodes. Additionally, the model exhibits a high potential for predictability over extended lead times, with a mean absolute error of 0.67 at 24 h, stabilizing at 0.75 at 72 h. This approach enhances simulation accuracy and provides policymakers extended response windows for O 3 control strategies. Highlights • A novel model by coupling attention scores and peak perception approach is developed. • Attention scores and peak perception approach improves hourly and peak O 3 prediction. • The wind direction, wind speed, and CO were the top three impact factors. • Transformer-IMW model displayed higher potential predictability for longer lead time.