Alternative Formulations and Improvements with Valid Inequalities for the Refinery Production Scheduling Problem Involving Operational Transitions

详细信息    查看全文

• 作者：Lu Zhang ; Yongheng Jiang ; Dexian Huang
• 刊名：Industrial & Engineering Chemistry Research
• 出版年：2015
• 出版时间：August 19, 2015
• 年：2015
• 卷：54
• 期：32
• 页码：7871-7889
• 全文大小：608K
• ISSN：1520-5045

文摘

Operation mode switching of production units would result in long transitions with fluctuant product yields, production costs, and key product properties. To formulate the remarkable process dynamics, a mixed-integer linear programming (MILP) model involving operational transitions of mode switching was proposed for the refinery production scheduling problem in Shi et al. ( Ind. Eng. Chem. Res. 2014, 53 (19), 8155鈭?170), which can describe transitional behaviors and provide implementable schedules. However, the model is very computationally expensive because it involves a large number of discrete and continuous variables, and the constraints about operational transitions are numerous and complex. In this paper, we study a scheduling problem similar to that in Shi et al. ( Ind. Eng. Chem. Res. 2014, 53 (19), 8155鈭?170) and aim at improving the computational efficiency of MILP models by alternative formulations and valid inequalities. First, we redefine some sets, variables and introduce new variables to propose the basic formulation, with which the nonlinear items are avoided and the variable definitions are simplified. Second, by reformulating constraints to explicitly describe the Fixed Charge Network Flow characteristic of operation mode switching, three Fixed-Charge-Network-Flow-based reformulations are proposed. Third, valid inequalities are developed for lot-sizing relaxations derived from the reformulations. Computational results show that the basic formulation is much smaller-sized and its computational performance is significantly better. The Fixed-Charge-Network-Flow-based reformulations together with the valid inequalities can further reduce by up to more than 95% the computational time while tightening the linear relaxation of the MILP model by more than 45%.