Kevin Tierney

A gender-based genetic algorithm for the automatic configuration of algorithms

A problem that is inherent to the development and efficient use of solvers is that of tuning parameters. The CP community has a long history of addressing this task automatically. We propose a robust, inherently parallel genetic algorithm for the problem of configuring solvers automatically. In order to cope with the high costs of evaluating the fitness of individuals, we introduce a gender separation whereby we apply different selection pressure on both genders. Experimental results on a selection of SAT solvers show significant performance and robustness gains over the current state-of-the-art in automatic algorithm configuration.

A mathematical model of inter-terminal transportation

We present a novel integer programming model for analyzing inter-terminal transportation (ITT) in new and expanding sea ports. ITT is the movement of containers between terminals (sea, rail or otherwise) within a port. ITT represents a significant source of delay for containers being transshipped, which costs ports money and affects a port’s reputation. Our model assists ports in analyzing the impact of new infrastructure, the placement of terminals, and ITT vehicle investments. We provide analysis of ITT at two ports, the port of Hamburg, Germany and the Maasvlakte 1 & 2 area of the port of Rotterdam, The Netherlands, in which we solve a vehicle flow combined with a multi-commodity container flow on a congestion based time–space graph to optimality. We introduce a two-step solution procedure that computes a relaxation of the overall ITT problem in order to find solutions faster. Our graph contains special structures to model the long term loading and unloading of vehicles, and our model is general enough to model a number of important real-world aspects of ITT, such as traffic congestion, penalized late container delivery, multiple ITT transportation modes, and port infrastructure modifications. We show that our model can scale to real-world sizes and provide ports with important information for their long term decision making.

On the complexity of container stowage planning problems

The optimization of container ship and depot operations embeds the k-shift problem, in which containers must be stowed in stacks such that at most k containers must be removed in order to reach containers below them. We first solve an open problem introduced by Avriel et al. (2000) by showing that changing from uncapacitated to capacitated stacks reduces the complexity of this problem from NP-complete to polynomial. We then examine the complexity of the current state-of-the-art abstraction of container ship stowage planning, wherein containers and slots are grouped together. To do this, we define the hatch overstow problem, in which a set of containers are placed on top of the hatches of a container ship such that the number of containers that are stowed on hatches that must be accessed is minimized. We show that this problem is NP-complete by a reduction from the set-covering problem, which means that even abstract formulation of container ship stowage planning is intractable.

Features for Exploiting Black-Box Optimization Problem Structure

Black-box optimization BBO problems arise in numerous scientific and engineering applications and are characterized by computationally intensive objective functions, which severely limit the number of evaluations that can be performed. We present a robust set of features that analyze the fitness landscape of BBO problems and show how an algorithm portfolio approach can exploit these general, problem independent, features and outperform the utilization of any single minimization search strategy. We test our methodology on data from the GECCO Workshop on BBO Benchmarking 2012, which contains 21 state-of-the-art solvers run on 24 well-established functions.

CP Methods for Scheduling and Routing with Time-Dependent Task Costs

A particularly difficult class of scheduling and routing problems involves an objective that is a sum of time-varying action costs, which increases the size and complexity of the problem. Solve-and-improve approaches, which find an initial solution for a simplified model and improve it using a cost function, and Mixed Integer Programming (MIP) are often used for solving such problems. However, Constraint Programming (CP), particularly with Lazy Clause Generation (LCG), has been found to be faster than MIP for some scheduling problems with time-varying action costs. In this paper, we compare CP and LCG against a solve-and-improve approach for two recently introduced problems in maritime logistics with time-varying action costs: the Liner Shipping Fleet Repositioning Problem (LSFRP) and the Bulk Port Cargo Throughput Optimisation Problem (BPCTOP). We present a novel CP model for the LSFRP, which is faster than all previous methods and outperforms a simplified automated planning model without time-varying costs. We show that a LCG solver is faster for solving the BPCTOP than a standard finite domain CP solver with a simplified model. We find that CP and LCG are effective methods for solving scheduling problems, and are worth investigating for other scheduling and routing problems that are currently being solved using MIP or solve-and-improve approaches.

The liner shipping fleet repositioning problem with cargo flows

We solve an important problem for the liner shipping industry called the Liner Shipping Fleet Repositioning Problem (LSFRP). The LSFRP poses a large financial burden on liner shipping firms. During repositioning, vessels are moved between services in a liner shipping network. Shippers wish to reposition vessels as cheaply as possible without disrupting the cargo flows of the network. The LSFRP is characterized by chains of interacting activities with a multi-commodity flow over paths defined by the activities chosen. Despite its great industrial importance, the LSFRP has received little attention in the literature. We introduce a novel mathematical model of the LSFRP with cargo flows based on a carefully constructed graph and evaluate it on real world data from our industrial collaborator.

A biased random-key genetic algorithm for the container pre-marshalling problem

The container pre-marshalling problem (CPMP) is performed at container terminals around the world to re-order containers so that they can be more efficiently transferred through the terminal. We introduce a novel decoder for a biased random-key genetic algorithm (BRKGA) that solves the CPMP. The decoder consists of a construction algorithm that learns how to best apply single and compound containers moves to quickly sort a bay of containers. Our approach finds better solutions than the state-of-the-art method on many instances of the standard pre-marshalling benchmarks in less computational time. Furthermore, we perform a computational analysis of different components of the BRKGA decoder to determine what types of heuristics work best for pre-marshalling problems, as well as conduct a feature space analysis of different pre-marshalling approaches. HighlightsA novel biased random-key genetic algorithm approach for solving the container pre-marshalling problem (CPMP).The first metaheuristic procedure that learns how to solve CPMP instances.Outperforms the state-of-the-art method on many instances of standard benchmarks.Provides a feature space analysis showing comparing state-of-the-art methods and on what types of instances they work best.

Solving real-world sized container pre-marshalling problems with an iterative deepening branch-and-bound algorithm

Abstract Container terminals around the world regularly re-sort the containers they store according to their retrieval times in a process called pre-marshalling, thus ensuring containers are efficiently transferred through the terminal. State-of-the-art algorithms struggle to find optimal solutions for real-world sized pre-marshalling problems. To this end, we introduce an improved exact algorithm using an iterative deepening branch and bound search, including a novel lower bound computation, a new branching heuristic, new dominance rule and a new greedy partial solution completion heuristic. Our approach finds optimal solutions for 161 more instances than the state-of-the-art algorithm on two well known, difficult pre-marshalling datasets, and solves all instances in three other datasets in just several seconds. Furthermore, we find optimal solutions for a majority of real-world sized instances, and feasible solutions with very low relaxation gaps on those instances where no optimal could be found.

An Algorithm Selection Benchmark of the Container Pre-marshalling Problem

We present an algorithm selection benchmark based on optimal search algorithms for solving the container pre-marshalling problem (CPMP), an NP-hard problem from the field of container terminal optimization. Novel features are introduced and then systematically expanded through the recently proposed approach of latent feature analysis. The CPMP benchmark is interesting, as it involves a homogeneous set of parameterized algorithms that nonetheless result in a diverse range of performances. We present computational results using a state-of-the-art portfolio technique, thus providing a baseline for the benchmark.

A Node Flow Model for the Inflexible Visitation Liner Shipping Fleet Repositioning Problem with Cargo Flows

We introduce a novel, node flow based mathematical model for the fixed-time version of a central problem in the liner shipping industry called the Liner Shipping Fleet Repositioning Problem (LSFRP). We call this version of the problem the Inflexible Visitation LSFRP (IVLSFRP). During repositioning, vessels are moved between routes in a liner shipping network. Shipping lines wish to reposition vessels as cheaply as possible without disrupting the cargo flows of the network. The LSFRP is characterized by chains of interacting activities with a multi-commodity flow over paths defined by the activities chosen. We introduce two versions of a node flow based model that exploit the fixed activity times of the IVLSFRP’s graph to handle cargo demands on the nodes of the graph, instead of the arcs, significantly reducing the number of variables. Using this model in CPLEX, we are able to solve 12 previously unsolved IVLSFRP instances to optimality. Additionally, we improve the solution time on every instance in the IVLSFRP dataset, sometimes by several orders of magnitude.