Research On Robust Optimization And Process Scheduling Under Uncertainty

In the process industry,production scheduling connects the upper-level long-term pro-duction planning and the lower-level real-time process control,which has been developed over the course of nearly half a century with a series of significant achievements.Howev-er,the uncertainty that widely exists in the production and business activities of process enterprises causes great challenges for its theoretical research and industrial practice.As a result,how to optimize the schedule such that the process can be operated safely and smoothly under uncertainty has been an important research topic.To this end,with a fo-cus on the robust scheduling method,this thesis addresses some of its pain points and bottlenecks that are exposed in practice using the adjustable robust optimization theory.After first systematically reviewing the process scheduling problem under uncertainty and the fundamental theory of robust optimization,the thesis carries out in-depth research on the method of Pareto optimal finite adaptability for robust scheduling problems,robust optimization under endogenous uncertainty,mixed-integer decision rule,distributionally robust scheduling methods and the connection between active learning and endogenous uncertainty,respectively.The main contents and contributions of this thesis are as follows:1.A novel adjustable robust optimization method based on finite adaptability is pro-posed for the production scheduling of an ethylene plant with uncertainties of energy con-sumption and decoking operations.To effectively address the time-varying and decoking-sensitive characteristics of uncertain fuel consumption parameters,the model of Pareto op-timal finite adaptability is built based on the uncertainty-set tree,which can yield a schedule that is not only Pareto optimal but also robustly optimal.It is demonstrated by the exten-sive simulation cases based on the industrial data that the proposed approach can adjust the schedule according to the observed uncertain decoking decision signal,which consequent-ly improves the cost-effectiveness of fuel pre-orders significantly without compromising the robustness.2.A multistage robust mixed-integer optimization approach is proposed for the en-dogenous uncertainty which widely exists in process scheduling problems.The method not only allows the continuous and binary recourse simultaneously,but also effectively models the uncertainty set affected by(recourse)decisions.The proposed mixed-integer decision rule based on the lifted uncertainty includes a type of discontinuous piecewise linear decision rule for the continuous recourse,which can realize the nontrivial integra-tion of decision-dependent uncertainty sets and recourse decisions.By doing so,tractable reformulations for the two-and multistage problems can be derived.The approach is ap-plied to a series of case studies,including a multiperiod production scheduling problem with endogenous uncertainty of production capacity,and the results show that the method can effectively address the endogenous uncertainty and significantly enhance the flexibility of schedule.3.Motivated by the coking uncertainty and corresponding decoking decisions of cracking furnaces,a multistage distributionally robust optimization approach is proposed for the integrated production and maintenance scheduling problem with degradation un-certainty.The data-driven Wasserstein ambiguity set is applied to capture the distributional uncertainty,and then the multistage distributionally robust scheduling model is built,which allows the mixed-integer recourse and optimizes the worst-case expected cost.To exam-ine its effectiveness and practicability,the approach is applied to extensive computational experiments,including a simulation case for the furnace system in a real-world ethylene plant.The results show that the proposed approach can lead to improved out-of-sample performance while maintaining computational tractability.4.The classification of endogenous uncertainty is further refined to distinguish the decision-dependent materialization and observation,and then the connection between ac-tive learning and adjustable robust optimization under endogenous uncertainty is highlight-ed.The decision-dependent nonanticipativity is effectively modeled with a set of auxiliary uncertain parameters,which results in a framework of multistage adjustable robust opti-mization that unifies the treatment of all types of endogenous uncertainty.The effectiveness and versatility of the proposed approach is demonstrated by computational experiments that cover a range of applications,including the integrated production and maintenance scheduling with inspections and the production scheduling with active parameter estima-tion.At the end of the thesis,promising future research interests in efficient algorithms for complex adjustable robust optimization problems,online robust scheduling methods as well as production scheduling and process control with active learning are discussed.