Simulation-based optimization in port logistics

Abstract

This is the summary of the author’s PhD thesis supervised by Pasquale Legato and defended on March 4, 2009 at the Universita della Calabria in Rende, Italy. The thesis is written in English and is available from rmazza@deis.unical.it upon request. Container terminal logistics have received great interest in the scientific literature from both the theoretical and practical standpoint. The reason for such concern is straightforward if one considers the number and randomnature of the operational activities carried-out in these facilities: vessel arrival and berthing, resource assignment and scheduling, container transfer and handling, emergency management (e.g. equipment failure, congestion phenomena, weather conditions) and so on. The common approach to problem design and solution in these logistic platforms is based on decomposing the original problem into several smaller models. However, to deliver overall system performance measures, a satisfactory contribution can spring from the use of an integrated methodology which can significantly aid decision-making under uncertainty: simulation-based optimization (SO). The SO methodological contribution proposed herein arises from considering the integration of stand-alone optimization techniques in discrete-event simulation, in order to first generate and then evaluate alternative solutions for managing dynamic logistic processes, under realistic conditions of uncertainty. As for solution generation, a metaheuristic approach based on a variant of the wellknown Simulated Annealing (SA) algorithm (see M.H. Alrefaei and S. Andradttir, “A simulated annealing algorithm with constant temperature for discrete stochastic optimization”, Management Science 45(5), 748–764, 1999) has been adopted. Besides discarding the basic assumption according to which the temperature Tk → 0 as k → 0 by assuming Tk = T ∀k, this approach bears two possible ways of estimating the optimum solution. It either uses the most visited solution or selects the solution with the best average estimated value of the objective function. The thesis first discusses the effectiveness of this constant temperature approachwhich