[Urban Near-surface Ozone Prediction Model Based on SARIMA-BP Neural Network]

With the acceleration of urbanization and industrialization, the problem of urban ozone (O₃) pollution in China has become increasingly serious. Aiming to address the limitation that the traditional time series model predicts O₃ concentration without fully considering the stochastic factors, a machine learning fusion model, i.e., the integrated model of seasonal autoregressive integral sliding average (SARIMA) and back-propagation neural network (BPNN), is proposed. The model decomposes the data into linear and nonlinear parts and fully utilizes the linear fitting advantage of the SARIMA model and the nonlinear mapping ability of the BPNN in order to improve the prediction accuracy. Specifically, the seasonal trend decomposition method (STL) was firstly applied to the original O₃ series to extract its trend, seasonal components, and random effects, based on which a SARIMA model was built to predict the linear changes in O₃ concentration. Subsequently, the nonlinear part of the data was input into the BPNN to fit the stochastic fluctuations. Ultimately, the prediction results of the SARIMA and the BP model were integrated to obtain the comprehensive prediction output. The O₃ concentration monitoring data of Hefei City from 2021 to 2023 were selected to construct a combined SARIMA-BP neural network model. The results showed that the root mean square error (RMSE) reached 8.385 2 μg·m^-3, which improved the prediction accuracy by 55.88% and 22.39% compared to that of the single SARIMA and BP models, and it was better than the SARIMA-LSTM model prediction effect, providing a theoretical basis for urban ozone pollution prevention and control.