PREDICTION OF PM_2.5 AND PM₁₀ CONCENTRATION IN GUANGZHOU BASED ON DEEP LEARNING MODEL

Precisely predicting the concentration of PM_2.5 and PM₁₀ in air pollution can provide a scientific basis for the prevention and control of air pollution. However, in the absence of pollution source emission inventory and visibility data, the prediction accuracy of the existing PM_2.5 and PM₁₀ concentration prediction methods are not high. In addition, it is rarely reported that the current deep learning models have been applied successively to PM_2.5 and PM₁₀ concentration prediction research. Based on the historical air quality monitoring data and weather monitoring historical data in Guangzhou from June 1, 2015 to January 10, 2018, two traditional machine learning models(random forest model(RF) and XGBoost model) and two deep learning models(short-long-term memory network(LSTM) and gated recurrent unit network(GRU) were constructed respectively, to predict the daily average concentration of PM_2.5 and PM₁₀ in Guangzhou. The results showed that the four models could also well predict the daily average concentration of PM_2.5 and PM₁₀ in the absence of pollution source emission inventory and visibility data. According to the evaluation metrics, i.e., MSE, RMSE, MAPE, MAE, and R², the PM_2.5 and PM₁₀ prediction effects of the four models were evaluated. The results indicated that the prediction effect of the deep GRU model was the best and the prediction results of the RF model were the worst. Compared with the commonly used RF model, XGBoost model, and LSTM model, the GRU model based on deep learning could better predict PM_2.5 and PM₁₀ concentration.