A Comparative Study On The Methods For Optimizing Batch Schedules Of A Multiproduct Pipeline Network

Most multiproduct pipelines adopt the back-to-back way to simultaneously transport several products.For ensuring the timeliness,security and economy of supplying products to markets,the essential task of operating a pipeline is drafting batch schedules in advance.A multiproduct pipeline usually comprises several input and delivery stations.During a given scheduling horizon,batch schedules of a pipeline must satisfy the limitations of delivery/inputting operations carried out by the initial station,intermediate stations and terminal station.Generally,drafting a feasible batch schedule of a pipeline is a hard work and optimizing batch schedules is more difficult.A multiproduct pipeline network usually contains several pipelines linked by transfer tank farms.The optimization of a pipeline network is more difficult than that of a single pipeline.For improving the efficiency and quality of the drafting work of batch schedules,this paper compares the performances of different approachs for optimizing batch scheules of a single pipeline and a pipeline network,respectively.Two mixed integer non-linear programming models for optimizing batch scheules of a single pipeline and a pipeline network are established,which are further transformed into mixed integer linear programming(MILP)models by mathematical conversion.The two MILP models aim to keep all pipeline segments of pipelines or pipeline networks to be the most stable.The feasibility of the objective function is proved based on the properties of convex function.A conclusion is drawn that if the flow rate of a pipeline segment is the most stable,the energy consumption for pumping products will be the lowest.For avoiding frequent change of delivery/injecting rates and frequent start-stop operations of operating a pipeline,the constraints about delivery/injecting rate stability of stations and minimum durations of every stoppage and every running of a pipeline are introduced.The two MILP models are solved using CPLEX.And results show that CPLEX is not suitable to solve large-scale MILP models.Then,the delivery operation relay(DOR)method and the parallel simulated annealing(SA)algorithm are proposed to optimize delivery schedules of a pipeline,respectively.DOR can generate high-quality delivery schedules in a short time,which is based on the heuristic rule that the relay of delivery operations carried out by adjacent stations can increase the flow rate stability of pipeline segments.Parallel SA can simultaneously generate several new solutions in every iteration,which can improve the efficiency of every iteration.Parallel SA adopts a two-stage framework to construct every new solution.The first stage uses the method of constructing a neighbourhood of a variable to adjust the old delivery schedule,which is further fine-tuned based on a heuristic rule about the proper relay of delivery operations in the second stage.The performance comparisons between the three methods for optimizing delivery schedules are made on three pipelines,which show that the rank is CPLEX,DOR and parallel SA for small-scale problems,and DOR,parallel SA and CPLEX for large-scale problems.For optimizing batch schedules of a pipeline network,an efficient method is proposed based on the framework of SA.The problem on optimizing batch schedules of every pipeline in a network is divided into a sub-problem on optimizing inputting schedules of initial stations and a sub-problem on optimizing delivery schedules of delivery stations.The second sub-problem is solved using DOR.Through decoupling treatment,the optimization method of batch schedules of a pipeline network is mainly focused on optimizing inputting schedules of initial stations of pipeline,which effectively reduces the computational burdens.Finally,the performance comparisons between SA and CPLEX are made on three pipeline networks with different structures,which show that SA not only consumes less computational time,but also achieves higher-quality solutions.