Evolutionary computation methods for the schedule optimisation of pipeline networks. (2016)

• Record Type: Journal Article
• Authors: Krause, Jonas; Lopes, Heitor Silvério
• Is Part Of: International journal of innovative computing and applications. Volume 7:Number 4(2016)
• Page Start: 191
• Page End: 201
• Publication Date: 2016
• Subjects: evolutionary computation -- differential evolution -- mixed integer linear programming -- MILP -- pipeline networks -- pipeline scheduling -- schedule optimisation -- oil pipelines -- mathematical modelling; Evolutionary computation -- Periodicals; Neural networks (Computer science) -- Periodicals; Genetic programming (Computer science) -- Periodicals; Biologically-inspired computing -- Periodicals; Swarm intelligence -- Periodicals; Quantum computers -- Periodicals; 006.3
• Languages: English
• ISSNs: 1751-648X

Abstract

Two evolutionary computation methods are presented in this paper, both variants of the differential evolution (DE) algorithm. Their main difference is the encoding process (binary and continuous) and both methods were successfully applied to the pipeline network schedule problem. A binary mathematical model is proposed to represent the flow of oil products in a 48 hours horizon period. This paper introduces new benchmarks of the pipeline scheduling problem for testing the proposed evolutionary algorithms on a specific network topology, but with different products and demands. Although computationally expensive, a mixed integer linear programming (MILP) approach is used to obtain optimal solutions so as to compare results with the evolutionary methods. MILP results achieved optimal solutions for nine out of the 15 benchmarks proposed, but it requires far more computational effort than the DE-variants. Even though it is a real-parameter algorithm, the DE can be considered as a good heuristic, which is an alternative for the discrete problem studied. The overall comparison of results between the proposed DE-variants and MILP supports the efficiency, robustness and convergence speed of DE algorithm suggesting its usefulness to real-world problems of limited complexity.