A Genetic Programming-Based Method to Derive Self-Optimizing Control Schemes for Dividing-Wall Columns

Self-optimizing control (SOC) aims to maintain controlled variables associated with system energy consumption at constant setpoints, thereby enabling near-optimal operation under various disturbance conditions. A core challenge in SOC scheme design is the selection of appropriate self-optimizing controlled variables (SOCVs). To address this challenge, a genetic programming (GP)-based method is proposed to identify linear combinations of process variables that minimize energy consumption, which enhances the scientific rigor and efficiency of SOCV selection. The proposed method is validated through a case study involving a dividing-wall column (DWC) for the separation of an ethanol-propanol-butanol ternary mixture. The derived SOC scheme incorporates three concentration-temperature cascade control loops to ensure the maintenance of product purities, alongside a temperature-inferential SOC loop dedicated to energy minimization. In this SOC loop, the liquid split ratio serves as the manipulated variable, while the GP-derived SOCV is a linear combination of temperatures from three sensitive stages. Closed-loop simulation results confirm that the proposed SOC scheme achieves stable and energy-efficient operation across multiple disturbance scenarios. Notably, compared with conventional control schemes featuring fixed liquid split ratios or single-temperature SOCVs, the proposed scheme eliminates the need for additional temperature measurements while realizing reduced energy consumption.