Predictive analysis of atmospheric sulfur dioxide concentration based on fusion modeling

Sulfur dioxide (SO2) is one of the most common components of atmospheric pollution, and its high concentration can cause harm to human beings as well as to the environment, and it is crucial to accurately predict the changes of sulfur dioxide content in the air. To this end, this paper first uses lasso regression and vector autoregressive model to do analysis of the factors affecting SO2 concentration, and then introduces the decomposability of NeuralProphet model for time-series data to decompose the SO2 concentration data into a trend term, a cyclic change term, and integrates it with other influencing factors to construct the Lightweight Gradient Boosting Machine (Lightgbm) model to obtain the SO2 concentration prediction value. The experiment is mainly analyzed by air pollution and meteorological history of Datong city, and compared with GA-BP and ARIMA, the results show that the R2 coefficient reaches 0.724, which is greatly improved compared with NeuralProphet, and the prediction accuracy of this model is better than that of ARIMA and GABP, and the factors affecting the concentration of SO2 are taken into account.