Application of Support Vector Machine Regression in Ozone Forecasting

Abstract

Support vector machine regression (SVMr) was proposed to forecast hourly ozone (O3) concentrations, daily maximum O3 concentrations, and maximum 8 h moving average O3 concentrations (O3 8 h) by employing the observations of meteorological variables and O3 and its precursors during the high O3 periods from May 20 to August 15, 2016 at an industrial area in Nanjing. The squared correlation coefficient (R2) of the hourly O3 concentrations forecast was 0.84. The mean absolute error (MAE) and mean absolute percentage error (MAPE) were 3.44×10-9 and 24.48, respectively. The key factors for the hourly O3 forecast were the O3 pre-concentrations, amount of ultraviolet radiation B (UVB), and the NO2 concentration. The main factors for the O3 daily maximum forecast were the NOx concentrations at 07:00 and the UVB level. Temperature and UVB played an important role in predicting O3 8 h. In general, taking precursors into account could increase the accuracy of O3 prediction by 10%-28%. For O3 concentration forecasting, SVMr gave significantly better predictions than multiple linear regression methods.