Choosing appropriate controlled variables(CVs)is demonstrated to be effective for real-time optimization(RTO)of chemical processes

whereas existing approaches do not allow the CVs to be changed once determined

which leads to limited RTO performances.To this end

this paper presents an approach for on-line modeling CVs.The local modeling method was used to select similar samples to construct estimating models for necessary conditions of optimality(NCO)

which were used as CVs to achieve better RTO performance under the functions of feedback controllers.A case study of evaporator demonstrated that the proposed method could increase the economic profit of process operation by accurate modeling for on-line NCO.