A Hybrid Deep Learning Model for Predicting PM2.5

Accurate prediction of PM
2.5
concentration is of great significance to air pollution control and human health protection. However, how to extract effective features of numerous information from a large amount of relevant monitoring data has become a challenging problem. In this paper, we propose a hybrid deep learning model (HDL) for predicting PM2.5 concentration. Firstly, we select four models that can process time-series data, including simple-RNN, LSTM, GRU, and TCN. These models have been trained as base models respectively with the air quality data and meteorological data in Changsha in the past two years. Secondly, the four base models are integrated into a new HDL model, which inherits all the advantages of the four base models. Finally, MAE, MAPE, MSE, RMSE, and R
²
are employed to evaluate the prediction accuracy of these models. The experimental results show that the HDL model has higher accuracy in most numerical scenarios than the above four models.