Forecasting Particulate Matter (PM 2.5) Concentration in Air with Double Attention GRU

Accurate PM2.5 concentration forecasting is crucial for reducing air pollution, protecting public health, directing urban development choices, and learning more about the effects of climate change. This research examines the Beijing multisite dataset to assess air quality based on PM2.5 concentrations. The air pollutant dataset is preprocessed. Feature engineering is carried out by utilizing a min-max scaler to handle the nonlinear correlation between attributes. The gated recurrent unit (GRU) with the attention mechanism has been exploited for PM2.5 concentration. GRUs excel in handling time series data based on their ability to capture long-term dependencies and temporal patterns, addressing the vanishing gradient issue. The integration of the attention mechanism is vital as it enables the model to concentrate on pertinent segments of the input sequence to enhance the predictive performance by capturing essential temporal dependencies and features. The model's generalization and reliability are assessed through multi-hour prediction with window sizes of 6, 8, and 12 and achieve the highest performance in terms of error measure RMSE, MAE, and R2score, with normalized and unnormalized datasets.