Prediction of pm2.5 concentration based on VMD-CNN-Transformer hybrid model

Abstract

In recent years, PM2.5 pollution has become increasingly serious, seriously affecting people's health and the world's ecological environment, and it has important research significance for the accurate prediction of PM2.5 concentration. In this paper, a VMD-CNN-Transformer hybrid model is proposed to predict future pm2.5 concentration. Firstly, the sliding window method is used to reconstruct the data set, and the required historical data range is set according to the window size. The research shows that the model has the best performance when the window size is 6. Secondly, the ablation experiment shows that the VMD module can enhance the predictive performance of the model. The test set is used to test the VMD-CNN-Transformer hybrid model, and the comparison experiment is conducted with the comparison model (LSTM, random forest). The results show that in terms of MAE, RMSE and determination coefficient, the Transformer model can be improved. The VMD-CNN-Transformer hybrid model can predict pm2.5 concentration better than the comparison model, with a determination coefficient of 0.97. Therefore, this study aims to improve the prediction of PM2.5 concentration, which is of great significance for controlling air pollution and solving health problems, and provides a scientific basis for related policies.