Modeling And Optimization Algorithm Of Planning And Scheduling In Petrochemical Industry

With the rapid development of the internet of things,artificial intelligence,digital twin and other technologies,intelligent manufacturing has built a technology system with the core of "data+model+algorithm+computing power" based on the end-to-end data flow in smart plants,which enables traditional manufacturing industry to improve quality and reduce con-sumption.Intelligent manufacturing has now become one of the core drivers for the digital and intelligent transformation of China's manufacturing industry.Among these technologies,intelligent decision-making is the core of intelligent manufacturing.The quality,reliability,timeliness and robustness of the decisions are important indicators of the level of enterprise intelligence.Multi-level and multi-system integrated decision-making,robust decision-making in uncertain environments and efficient optimization algorithms of large-scale decision-making problems are the hot research topics in intelligent decision-makingThe shale gas field development design and planning problem and the production plan-ning and scheduling problem,respectively,in the upstream and midstream or downstream of the petrochemical supply chain,are studied in this thesis.The modeling of design and planning integration,production planning and scheduling integration in the petrochemical industry under typical uncertainties,and the decomposition method for these Mixed-Integer Linear Program-ming(MILP)models are the main topics that are researched in this thesis.Besides these,this thesis also looks into the general decomposition methods and solver technologies for Mixed-Integer Nonlinear Programming,which is widely applied in the petrochemical industry,such as crude oil operation scheduling and process design.Overall,our research tries to improve the intelligent level of enterprise-wide decision-making from three aspects,the integrated modeling of complex systems,optimization under uncertainty and efficient optimization algorithmsThis thesis is organized as follows·To handle shale gas field development design and planning problem in the upstream of petrochemical supply chain,a large-scale mixed-integer linear programming model is proposed based on the shale gas field superstructure.The decisions in this problem in-clude the selection of candidate locations of wells and pads,the production operation scheduling,gas pipeline installation and pipeline size selection in shale gas field devel-opment.Considering the limitations of traditional methods and solvers,a solution-pool-based bilevel decomposition algorithm is proposed to solve the model efficiently after analyzing the model structure.The case study of five different examples demonstrates the advantage of the proposed model and the bilevel decomposition algorithm.The proposed model improves the economic benefits of shale gas development projects by making in-tegrated decisions on shale gas network design and shale gas field development planning in this problem.·A multistage stochastic programming model is proposed to address the shale gas field de-velopment planning problem under production profile uncertainty.The production profile uncertainty of shale gas wells belongs to the type 2 endogenous uncertainty and its re-alization is dependent on development decisions.Generalized Disjunctive Programming is applied to model the logical relationship between decision variables and uncertain pa-rameters.The resolution delay of uncertain parameters is also considered in the model.Based on the model structure of the multistage stochastic programming model under en-dogenous uncertainty,Lagrangean decomposition method and the heuristic strategy are applied to solve the model.The case study illustrates that the proposed model can ef-fectively reduce the risk of developing low production wells through planning the proper development sequence of candidate wells.·To address the production planning and scheduling problem under demand uncertainty in the midstream and downstream petrochemical supply chain,a modeling framework of planning and scheduling integration is proposed based on multistage stochastic program-ming.In the framework,the planning level and scheduling level are connected by the coupling constraints in decision granularity and time scale.The demand uncertainty is characterized by a scenario tree.To effectively solve the proposed model,a progressive hedging algorithm with multiple accelerating strategies is proposed.The case study of the state task network(STN)and a real-world ethylene plant demonstrates the advantage of the proposed model and algorithm,which resolve the issue of production benefit decline caused by fractional decision-making and intractability of uncertainties.·In enterprise-wide optimization in the petrochemical industry,many problems can be modeled as Mixed-Integer Nonlinear Programs.Outer Approximation method is an ef-ficient algorithm for solving MINLP problems.In order to guarantee the global conver-gence of the Outer Approximation method in nonconvex MINLP problems,McCormick relaxation-based global Outer Approximation method and global LP/NLP branch and bound algorithm are proposed.Valid cutting planes are generated based on McCormick convex relaxation and concave relaxation of nonconvex constraints.The polyhedral ap-proximation of the feasible region of the nonconvex MINLP problem can be further con-structed.The global convergence of the two algorithms are guaranteed by adding integer cuts or tabu list.The test of the crude-oil operations scheduling problem and hundreds of numerical and engineering instances demonstrates the excellent performance of the proposed algorithm for solving nonconvex MINLP problems·To prevent the big jump of the incumbent solution between adjacent iterations in Outer Approximation method,a regularized Outer Approximation method is proposed.By solv-ing the regularization problem,the optimal solution of the master problem is projected into the trust region and the moving range of the optimal solution of the master problem is then limited.The regularization problem generates an equivalent trust-region constraint by limiting the boundary of the objective function.l1 norm,l2 norm,l? norm,the first-order and second-order approximation of Lagrange function based trust region method are proposed for this framework.The benchmark of hundreds of open-source numerical cases and engineering instances shows that regularization can effectively reduce the num-ber of infeasible integer combinations and reduce the number of iterations required for convergence,demonstrating the superiority of regularized Outer Approximation method.At the end of the thesis,promising futures researches on integrated modeling framework of enterprise-wide optimization under endogenous and exogenous uncertainty,decomposition methods for Mixed-Integer Programming are discussed.