Novel MILP Decomposition Approach for Scheduling Product Distribution through a Pipeline Network

This work presents an approach for scheduling of operational activities in a large real-world pipeline network, where oil derivatives and ethanol are transported and distributed among refineries, terminals, depots, and final clients. The hierarchical decomposition approaches to solve the pipeline-scheduling problem presented by Boschetto et al. [Ind. Eng. Chem. Res. 2010, 49, 5661] and Magatao et al. [Ind. Eng. Chem. Res. 2012, 51, 4591], which are based on the integration of mixed integer linear programming (MILP) models and a set of heuristic modules, are merged and compounding blocks are also improved. Thus, a novel decomposition approach for scheduling product distribution through a pipeline network is proposed. In addition, this work presents a new MILP approach for the last hierarchical level: the timing block (timing model). This paper expands and improves the former MILP model, which was the timing block core. A series of operational constraints were considered within a continuous time representation in order to determine the exact time instants that products should be pumped into the pipelines and received in the operational areas during a scheduling horizon of, typically, 1 month. Within the new MILP timing model, turn shift constraints, local constraints, and surge tank constraints are improved; immediate pumping constraints are proposed. In addition, a decomposition approach for the new MILP model is also proposed within this article. This decomposition is based on a relax-and-fix heuristic implemented by a sequential run of two MILP models: MLC (Model with Local Constraints) and MST (Model with Seasonal costs and Turn shift constraints). The MILP decomposition goal is to reduce the computational load, if seasonal costs and turn shift constraints are active, without quality solution losses. The proposed approach is applied to the solution of real case studies of a pipeline network that includes 30 bidirectional multiproduct pipelines associated with 14 nodes (four refineries, two harbors, six depots, and two final clients). Computational results have been attained in a reasonable computational time (from seconds to a few minutes) for the addressed pipeline network.