Air Quality Time Series Analysis and Prediction Based on SMA-Transformer-QR

Traditional air quality prediction methods often show limitations when faced with complex and dynamic environmental conditions, especially in accurately predicting extreme weather events and environmental pollution. To address the challenges posed by the highly variable nature of the environment, this study develops an air quality prediction model based on SMA-Transformer-QR. First, the Transformer model is introduced, leveraging its advantage in capturing long-term dependencies in time series data to more accurately model the dynamic changes in air quality indicators over time. Next, quantile regression is employed as the loss function, with the SMA slime mold algorithm used to perform fine-tuning and optimization of the model's hyperparameters. A baseline comparison with five other models (MLP, RNN, GRU, LSTM, and Transformer) demonstrates that the Transformer-QR model achieves an average improvement of 23.3% over traditional models in terms of MAE, RMSE, MAPE, and $\mathrm{R}^{2}$ evaluation metrics. The SMA-Transformer-QR air quality prediction model established in this study can accurately forecast air quality under complex and dynamic environmental conditions, providing a reference for air quality management and prediction.