一类包含两种变量的调度优化问题理论方法研究

As the demand of energy saving increasing, the advanced process control (APC) and optimization technology (OT) have been developed on refinery device. Since a large part of crude oil in China is imported from different place all over the world, the qualities are far different from each other, this make APC and OT difficult to implement. The schedule optimization of crude oil blending (SOCOB) is the necessary method to accomplish refining plan and keep the blent oil qualities stable. The SOCOB problem is a large scale complex non-convex non-linear hybrid-varariable optimization problem, and is time consuming to solve. The problem contains sequence and continuous variables with some certain hiberarchy logic, and is the reprsentive of a class of optimization problem. The SOCOB problem is transformed to a two-layer problem according to its logic character, and the sequence variables are selected as decision variables, the optimization of continuous variables are taken as evaluation of the relative sequence variable in the proposal. The whole problem is studied with ordinal optimization ideas. The solution scheme including evolution mechanism and the balance between search width and evaluation depth, the stochastic evaluation including evaluation convergency model and data fitting, the constrained ordinal optimization will be studied in order to establish a kind of new solution scheme, solve some theoretical problems,design efficient algorithm, support implementation of APC and OT, drive the development of the schedule optimization problem with two kind of variables.

随着节能降耗要求的提高，炼油装置先进控制和优化技术不断推广。由于我国原油大量需要进口，来源分布广，性质差别大，导致先进控制和优化技术难以实施。原油调合调度优化是在原油性质多变的条件下完成炼制计划、保证常减压装置进料性质稳定的必要手段。原油调合调度优化是一个大规模复杂非凸非线性混合变量规划问题，其求解非常耗时。此问题包含顺序变量和连续变量，二者具有一定的层次逻辑，其结构代表了一类包含两种变量的优化问题。本课题将问题按逻辑关系分层，以顺序变量为优化变量，以优化连续变量为顺序变量的评价，以序优化思想研究优化算法。研究以进化机制和求解效率与评价深度和搜索广度的关系为主要内容的求解框架、包括概率评价收敛趋势模型和数据拟合的顺序变量概率评价以及约束序优化问题，以形成一种新的求解框架，解决若干基础理论问题，设计有效的求解算法，支持先进控制和优化技术的实施，推动一类含有两种变量的调度优化问题的研究与应用。

课题以典型流程工业为主的对象，研究了其过程模型和调度模型的建立问题，提出了装置级先进控制和厂级调度优化的集成策略，研究了基于分片线性、深度学习等的过程建模方法。课题在研究炼油过程调度优化建模的基础上，研究了基于FCNF和VI的模型重构问题，建立了高效率的模型。课题进一步研究了针对两层结构模型的求解算法：提出了基于拉格朗日分解的两层算法，并通过概率次梯度、拉格朗日乘子初始化和结构变换提高了求解效率；提出了两层混和优化算法；针对聚氯乙烯过程提出了基于结构分解的两层优化策略和算法；在算法中通过平衡子问题求解与迭代过程来提高效率。针对轨迹规划问题，建立了轨迹预测概率模型和集成换道决策和轨迹规划的数学规划模型，提出了一种基于序优化的求解方法，并设计了基于计算量分配的改进算法。这些建模和优化方法有效解决了相关问题，为更多的两层结构问题的解决提供了借鉴。