A mixed-integer optimization strategy for oil supply in distribution complexes

As a result of an increasingly competitive market, companies must find ways to organize their activities regarding their economic outcome. An important feature in this context involves transportation operations, usually considered one of the major bottlenecks in the production chain. While delays imply loss of time and lack of resources, deliveries ahead of the deadlines may cause excess of inventories. Therefore, every company must pursue efficient transportation schedules within their operational planning. This work addresses short-term crude oil scheduling problems in a distribution complex that contains ports, refineries and a pipeline infrastructure capable of transferring oil from the former to the latter. The ports comprise piers, which receive vessels for discharging, storage tanks and a network that connects each other. The refineries have their own storage infrastructure, modeled as a large storage unit, along with crude distillation units, considered as constant level consumers. The problem involves a number of other issues, including intermediate storage, settling tasks and allocation of crude oil by its qualitative characteristics. A decomposition strategy based on large-scale mixed-integer linear programming (MILP) continuous-time models is developed. First, an MILP model that considers an aggregate representation for the pipeline and intermediate storage infrastructure is proposed. Decision variables involve the assignment of oil tankers to piers as well as tanker unloading and pipeline loading operations. The solution of this model provides the initial conditions for MILP models that represent the pipeline and intermediate storage infrastructure at a detailed level. Algorithms based on the LP-based branch-and-bound method are employed. Results from a port scenario of 13 tankers, 4 piers, 14 crude types, 18 storage tanks and 2 pipelines were obtained in approximately 90 minutes from an MILP containing 1996 continuous variables, 1039 binary variables and 7203 constraints.