Based on tropical cyclone (TC) data, reanalysis data, and ozone monitoring data from 2014 to 2024, the synoptic mechanism by which approaching tropical cyclones (ATCs) trigger ozone pollution in the Guangdong-Hong Kong-Macao Greater Bay Area (GBA) was analyzed. A prediction model was developed using Spearman's rank correlation coefficient and the random forest regression method. The results indicated that: ATCs inducing ozone pollution in the GBA were primarily distributed within a range of 270~1758km, and a clear disparity in intensity was observed on the eastern and western sides of the line connecting Taiwan Island and Luzon Island. The ATCs east of the line all reach typhoon intensity or above, while those west range from tropical depression to severe typhoon intensity. Ozone pollution was induced by ATCs through the generation of subsiding airflows in upper levels, which reduces cloud cover over the GBA, increases downward shortwave radiation, elevates surface temperatures, and reduces relative humidity, precipitation and wind speed. Through SOM clustering analysis, four ozone pollution distribution patterns triggered by ATCs in the GBA were identified: the southwestern pollution pattern, the widespread pollution pattern, the western pollution pattern, and the northeastern pollution pattern. These patterns are associated with the transport and convergence effects of different surface wind fields. Ozone concentrations in the 11cities of the GBA were found to correlate well with tropical cyclone characteristics (radius of 7-level wind circle and proximity rate), as well as significant correlations with surface meteorological factors (such as land-sea wind direction difference and land-sea meridional wind speed difference) and upper-air meteorological factors (such as 200 hPa vertical velocity, 700 hPa meridional wind, 700 hPa specific humidity, and 925 hPa zonal wind). The forecasting model constructed based on 18factors, including tropical cyclone characteristics and surface and upper-air meteorological elements, demonstrates high generalizability and accuracy. The average correlation coefficient between the model's forecast values and the actual monitoring values reached 0.811, the average root mean square error was 30.2μg/m³, and the average TS score for ozone level prediction was 0.668.