Optimizing the ENAP's systems for distributing refined petroleum products by tankers and pipelines

Abstract

This thesis is compused by two research projects developed at Empresa Nacional del petroleo, ENAP. In the first part of this thesis we present an optimization framework for determining the optimal schedule for a fleet of tankers delivering refined petroleum products for ENAP, a state-owned company in Chile. The scheduling process addresses the distribution task of satisfying the demand for multiple products to multiple clients while minimizing the overall cost. These clients have different storage capacities and are located near ports along the coast. The optimization model is an exact one, based on defining each leg of each trip for each tanker. It was however a very large scale MIP model unsolvable using a commercial code (Node Model). We generated an approximate model based on a set of feasible routes for each tanker. This model make the problem more tractable and provides good feasible solutions (Routes model). The best solution is used as a hot start for the original Node Model, allowing its solution with very low residual gaps in moderate computer times. This two-step approach was implemented and has been used at ENAP since 2015. A graphical user interface provides a friendly format for the schedulers. The new system replaces the manual process that was used in the past. It has been a milestone change at ENAP and it is currently saving 8-9% of the total operational cost for the maritime distribution system. In the second part of this work we describe the creation, development, implementation and impact of a system optimizing the distribution of petroleum products by pipeline at ENAP. Scheduling used to be done by hand, based on experience and on the need to satisfy physical constraints in the pipeline, supply constraints in and out of the refinery and demand constraints from clients. The complexity of the problem and the need to cut down on operating costs suggested turning to optimization, specifically integer programming. The positive results of the project owe a lot to the constant interaction between schedulers, deciders, and the optimization team, and to the insights provided by the schedulers that allowed to limit the model's complexity. The resulting system is easy to use for schedulers thanks to a graphical user interface (GUI), and its solution requires little computer time. It is used once a month for planning the next month operations and negotiating delivery dates and amounts with the clients based on the solution suggested by the model. Operating cost savings are of the order of 10%. The system has also been used to evaluate an alternative distribution approach.