Based on training samples of a numerical simulation model established for boiler combustion system, a least squaresupport vector machine (LSSVM) model was set up, which was subsequently corrected with experimental data and updated with realtime operation data. On above basis, a hybrid modeling method was finally put forward for boiler combustion system. Results show that the hybrid modeling method could effectively improve the prediction precision and calculation speed of original LSSVM model, and by correcting and updating the model respectively with experimental data and realtime operation data, the model precision could be further improved, thus laying a good foundation for optimization of boiler combustion systems.