Abstract: NO2 is one of the main atmospheric pollutants, which plays an important role in atmospheric photochemical process. It is of great significance to study the temporal and spatial variation law of NO2 concentration and predict the variation trend of NO2 concentration. The BP neural network based on particle swarm optimization (PSO) was proposed to predict atmospheric NO2 concentration. Based on the air pollution data and meteorological data of Hefei area, China, from January 1, 2017 to December 31, 2019 and combined with the stepwise regression method, the influencing factors with high correlation with NO2 concentration were selected as the input samples. The PSO-BP neural network prediction model was constructed, and then the optimal solution of the initial weight and threshold value of the BP neural network were found by using PSO algorithm. By comparing the prediction results of the traditional BP neural network, BP neural network improved by genetic algorithm and BP neural network improved by PSO, it was found that PSO-BP model can accurately predict the dynamic change of NO2 concentration with high prediction accuracy and simple model, which is expected to be widely used in air pollutant concentration prediction in the future.

Key words: particle swarm optimization, back propagation neural network, stepwise regression, NO2 concentration prediction