Abstract: Using ozone monitoring data and meteorological data in Hohhot city from 2018 to 2022,this study analyzes the O₃ concentration variation characteristics.Models with relating meteorological factors to predict daily maximum 8-h average O₃ concentration (ρ(O₃-8h)) were established using Light Gradient Boosting Machine (Light-GBM),Extreme Gradient Boosting (XGBoost),Random Forest (RF),and Long Short-Term Memory networks (LSTM).A comparison of model performance metrics was conducted to identify the optimal model,which was validated.The results showed that over the past five years,O₃ concentration in Hohhot City (ρ(O₃-8h) exceeded 160 μg·m^-3 from April to October,with the most exceedance days in June and July,followed by May and August.Among the model input factors,daily maximum temperature contributes the most to the prediction of (ρ(O₃-8h),accounting for 44%.The LightGBM model provided the best simulation results,with overall model performance ranking from highest to lowest as follows: LightGBM> LSTM> XGBoost> RF.The locally constructed ozone meteorological condition assessment index showed a correlation coefficient up to 0.86 with (ρ(O₃-8h),an improvement of 17.81% over the China Meteorological Administration's ozone meteorological condition assessment index.This demonstrates its effectiveness in assessing meteorological condition influences on O₃ concentration variations in Hohhot City.

Key words: Extreme Gradient Boosting (XGBoost), Random Forest (RF), Long Short-Term Memory networks (LSTM), Light Gradient Boosting Machine (LightGBM)