In process industries, there is a growing trend toward low carbon society. Especially in oil refinement and petrochemical industries, furnaces consume a lot of energy. In order to reduce the energy consumption, excess O2 has to be reduced safely. However, high gain O2 control cannot be done due to L/T > 1. Therefore, conventional O2 controllers are operated with low gains, that is, high stable control performance for disturbances has not been able to obtain, then minimum excess O2 control cannot be achieved. The proposed method that utilizes the estimated O2 variable can derive much better control performance. Moreover, since the proposed estimation technique consists of a physical model part and a stochastic soft‐sensor, both estimation accuracy and reliability for extrapolation can be obtained. The validity of the proposed scheme is verified through actual application results.