Self-optimizing control methodology for mixed integer programming problems: a case study of refinery production scheduling

Abstract

Problem formulation as mixed integer nonlinear programming (MINLP) is one of the most challenging task in refinery scheduling optimization. In most of the work reported in refinery scheduling, uncertainties from design point of view predominate. However, there is also a need to consider operational uncertainties (disturbances) as they affect the accuracy and robustness of the overall schedule. This study proposed a novel approach under self- optimizing control (SOC) framework to deal with multi-period refinery scheduling problems under uncertain conditions. The goal is to maintain global optimum by controlling the gradient of the cost function at zero via approximating necessary conditions of optimality (NCO) over the whole uncertain parameter space. A regression model for the plant expected revenue (profit) as a function of independent variables using optimal operation data was obtained and a feedback input (manipulated variable) was derived. The performance of the proposed approach was tested using case studies. The first case assumed a system with no disturbance with the base case model giving an optimal profit of $56,696,407 while the proposed approach yields $50,523,054, translating to 10.888 % loss. The percentage loss for the second, third and fourth cases with disturbances are 5.807 %, 4.409% and 7.898% respectively. The results obtained have shown that the idea presented was able to effectively deal with the situation at hand with percentage loss within a reasonable degree

Keywords: Refinery scheduling scheduling, MINLP formulation, Operational uncertainty (disturbances), Necessary condition of optimality, Feedback control