炼油厂运营优化问题是原油产业链中非常重要的问题, 在学术界和工业界都有非常多的研究和应用。一般而言, 炼油厂优化问题会被建模为混合非线性整数规划问题(MINLP)。由于原油品种和相关产品繁多并且加工装置复杂多样, 所以变量维度较大。并且在具体的加工过程中涉及物料物性变化和加工规则, 从而产生非凸非线性和整数约束, 使得问题求解难度变大。目前学术界研究主要针对小规模问题或者运营流程的子系统进行建模求解, 并且求解方法集中在使用商用求解器, 如GAMS环境中的BARON、DICOPT等。本文针对炼厂优化的MINLP提出了一种混合分布递归和分支定界算法(Hybrid-DRBB), 分别对非线性约束和整数约束进行松弛和求解, 从而得到原问题的近似最优解。在实际的工业场景大规模数据中, 本文的算法速度被证实优于直接调用求解器的建模求解方式。

The optimization of refinery operations is a critical issue within the crude oil supply chain, garnering extensive research and application interest across both academic and industrial sectors. Typically, refinery optimization problems are modeled as Mixed Integer Non-Linear Programming (MINLP) problems. The complexity of these problems arises from the vast diversity of crude oil types and their derivative products, coupled with intricate processing procedures. Moreover, specific processing steps involve changes in material properties and rules, introducing non-convex, nonlinear, and integer constraints, which increase the solution-finding difficulty. Currently, academic research primarily focuses on modeling and solving small-scale issues or subsystems of operational processes. Commercial solvers like BARON and DICOPT in GAMS are commonly employed for such tasks. This paper introduces a Hybrid Distribution Recursion and Branch and Bound algorithm (Hybrid-DRBB) for solving MINLP in refinery optimization. This method relaxes and solves nonlinear and integer constraints separately, thereby obtaining near-optimal solutions for the original problem. Through numerical experiments utilizing real-world, large-scale data scenarios, we demonstrate the efficiency of our method in comparison to traditional commercial solvers, highlighting its reduced computational cost and improved solving approach.