蒸馏与换热协同的约束多目标在线操作优化方法

针对蒸馏装置与换热网络间缺乏协同优化导致的分馏精度差和能耗高的问题，提出了一种基于代理模型的约束多目标在线协同操作优化方法.为了解决蒸馏装置与换热网络操作参数协同优化时存在的计算耗时和约束的问题，构建Kriging代理模型来近似目标函数和约束条件，提出了基于随机欠采样和Adaboost的分类代理模型（RUSBoost）来解决类别不平衡的收敛判定预测问题.提出了基于多阶段自适应约束处理的代理模型的模型管理方法，该方法采用基于参考向量激活状态的最大化改善期望准则和可行概率准则更新机制来平衡优化初始阶段种群的多样性和可行性，采用支配参考点的置信下限准则更新机制加快收敛速度.通过不断与机理模型交互来在线更新代理模型，实现在线操作优化.通过测试函数和仿真实例验证了本文方法的有效性.

A Constrained Multi-objective Online Operation Optimization Method of Collaborative Distillation and Heat Exchanger Network

To solve low separation precision and high energy-consuming caused by lacking of collaborative operation optimization between atmospheric distillation unit and heat exchanger network, this paper presents a constrained multi-objective online collaborative operation optimization method based on surrogates model. To solve the problems of time-consuming and constraints in operating parameters collaborative optimization of distillation unit and heat exchanger network, the Kriging surrogate models are built to approximate each objective function and constraint, a classification surrogate model based on random undersampling and Adaboost is presented to solve the class imbalance of convergence prediction problem. A model management method of surrogate models based on multi-stage adaptive constraint handling is proposed. The method uses the maximization expected improvement and probability of feasibility criterion updating mechanism based on the reference vector activation state to balance the diversity and feasibility of the population at the early stage of evolution. The convergence rate is accelerated by using lower confidence bound criterion update mechanism dominating the reference point. By constantly interacting with the mechanism model to online update the surrogate model, the online operation optimization is realized. The efficiency of the proposed algorithm is validated by the results of benchmark functions and simulation example.