多技术集成的炼油过程短期生产计划优化新方法

Oil refinery industry plays a vitally important role in the national economy. With low productivity, high energy consumation, high cost, and low rate of resource usage, the optimization of short-term scheduling for oil refinery can gain great profit and increase competativeness. However, the oil refinery process is a hybrid system and feasibility is essential for its short-term scheduling problem. Hence, heuristics and computational intelligent techniques, such as genetic algorithm, are unable to apply. Thus, researchers have to formulate the problem as mixed integer programming problem and solve the problem by a exact way. Because of the NP-hard nature of the problem, such methods cannot overcome the barrier of computational complexity. To overcome this problem, this project proposes a novel approach by integrating the discrete event system control theory and various optimization techniques. By this methodology, the process is modeled by a hybrid Petri net and schedulability is studied based on the model. With the schedulability conditions obtained, the problem can be decomposed into two levels. At the upper level, it dedicates to the refining schedule, while, at the lower level, it finds the detailed schedule. Under such architecture, the short-term scheduling problem can be solved by solving a number of small sub-problems with discrete event variables or continuous variables only, but not both. In this way, it decouples the discrete event and continuous variables and overcomes the computational complexity problem resulting from the hybrid properties. Meanwhile, different objectives can be associated with different sub-problems. Thus, a breakthrough in this field is expected.

炼油工业在国民经济中具有及其重要地位。但它生产率低、能耗高、成本高、资源利用率低。因此，短期生产计划的优化可以带来巨大的效益和提高竞争力。但由于它的混合系统特征和解的可行性问题，启发式方法和像遗传算法这样的智能方法无法应用，计算复杂性成为问题的关键。为此，现有的研究用混合整数规划建模，并用精确算法求解。但是，由于问题的NP-hard本质，数学规划无法逾越计算复杂性的难点。本项目提出一种将离散事件系统控制理论和各种优化技术相结合的新方法。该方法建立混合Petri网模型，利用模型进行可调度性分析，获得可调度的条件。在此基础上将问题分解为上层的蒸馏塔炼油计划优化和下层的详细计划优化。在这一研究框架下可以将问题分解为单纯的离散优化和单纯的连续优化子问题，克服由于离散事件变量与连续变量交叉而带来的计算复杂性，同时不同优化指标在不同子问题中进行优化，从而解决计算复杂性难题。因此，有望获得本领域的突破。

炼油工业在国民经济中具有及其重要地位。短期生产计划的优化可以带来巨大的效益和提高竞争力。但由于它的混合系统特征和解的可行性问题，启发式方法和像遗传算法这样的智能方法无法应用，计算复杂性成为问题的关键。为此，现有的研究用混合整数规划建模，并用精确算法求解。但是，由于问题的NP-hard本质，数学规划无法逾越计算复杂性的难点。本项目提出一种将离散事件系统控制理论和各种优化技术相结合的新方法。在这一框架下，对不同系统构成在不同运行条件下建立混合Petri网模型，利用模型利用控制理论原理进行可调度性分析，获得可调度的条件。在此基础上将问题分解为上层的蒸馏塔炼油计划优化和下层的详细计划优化。然后，对上层蒸馏塔炼油计划优化问题进一步分解实现离散事件变量与连续变量的解耦。结果，可以基于线性规划的方法求解。对于下层详细计划优化问题，获得了简单的递推方法而求得可行解。进一步，首次成功地将遗传算法(GA)用于求解这一问题。为此，本项目解决了计算复杂性的难题，获得的方法可以用于开发实际可用的软件工具。这是本领域的重要突破。发表了论文25篇，其中13篇发表在本领域重要的SCI索引国际期刊上，7篇已正式发表，6篇也已在线发表。