Air Pollution Level Prediction Based on VMD-DBO-LSTM-ATT

The prediction of air pollution levels is crucial for improving people's quality of life. To tackle the challenges of air pollution variability and low prediction accuracy, a prediction model is proposed that combines Variational Mode Decomposition (VMD) and Dung Beetle Optimization (DBO) with an optimized Long Short-Term Memory (LSTM) network. Initially, to minimize the effect of noise on the prediction results, the VMD method decomposes the non-stationary AQI data, extracting modal components with distinct features. Next, to address the limitations of manually adjusting LSTM parameters, the DBO algorithm is applied to optimize the model's parameters. The attention mechanism (ATT) is then incorporated to assign weights to significant data, enhancing the model's accuracy. Lastly, the individual sub-sequences are merged to produce the final prediction. The inclusion of the attention mechanism demonstrates that the VMD-DBO-LSTM-ATT model outperforms other models, with the coefficient of determination (R-square, R2) improving by 8.3%.