Local search based methods for scheduling in the unrelated parallel machines environment

https://doi.org/10.1016/j.eswa.2022.116909Get rights and content

Highlights

  • Initial solution selection has high influence on metaheuristic performance.

  • Simple local search methods can easily outperform metaheuristics.

  • Simple metaheuristics perform better than more complex ones.

  • The selection of appropriate local search operators is mandatory.

  • More effort should be invested in local search than metaheuristic design.

Abstract

In many real-world situations it is necessary to make timely scheduling decisions. In most cases, metaheuristic algorithms are used to solve various scheduling problems because of their flexibility and their ability to produce satisfactory results in a short time. In recent years, several novel or hybrid metaheuristics have been proposed for scheduling problems. Although such research leads to new insights, it inevitably causes certain problems. First, it becomes unclear which methods perform best, especially if they are not properly compared with existing ones. Second, the proposed methods become increasingly complex, making them more difficult to understand and apply. The goal of this study is to investigate the possibility of defining efficient but simple iterative local search (ILS) methods for the parallel unrelated machines environment with minimisation of the total weighted tardiness. To improve the efficiency of ILS methods, several design decisions, such as the construction of the initial solution and choice of local search operators. The proposed methods have been compared with several metaheuristics, of which they achieve significantly better results. Thus, we conclude that it is not necessary to increase the complexity of metaheuristics to achieve better results. Rather, better results can be obtained with simple but well-designed local search methods.

Introduction

Scheduling processes are important in many real-world situations, such as planning in manufacturing plants (Kofler et al., 2009, Ouelhadj and Petrovic, 2009), universities and schools (Lewis, Paechter, & Rossi-Doria, 2007), airports (Cheng et al., 1999, Hansen, 2004), hospitals (Burke et al., 2004, Petrovic and Castro, 2011), and many others. In such problems, the goal is to assign a certain number of jobs or tasks to a limited number of machines or resources in such a way that all required constraints are satisfied and a certain user-defined criterion is optimised (Pinedo, 2012). Solving most scheduling problems is difficult because they are NP hard. Exact methods can provide optimal solutions to such problems (Fanjul-Peyro, Ruiz, & Perea, 2019), but they cannot be applied to large-scale problems because it is not possible to enumerate the entire search space. Therefore, most research focused on the development and use of methods that do not necessarily provide optimal results, but can achieve good solutions in a reasonable amount of time. The studies mainly focused on two types of methods, approximation methods (Lenstra, Shmoys, & Tardos, 1990) and heuristic methods (Morton & Pentico, 1993). Approximation methods provide some guarantee of optimality of the solution (Pinedo, 2012, Wotzlaw, 2012), while heuristic methods provide no guarantees.

Of the previous two types of methods, heuristic algorithms have attracted considerable attention because they can be adapted to a wider range of problems and are easier to design. Heuristic methods used for solving scheduling problems can be divided into problem-specific heuristics and metaheuristics. Problem-specific heuristics usually appear in the form of dispatching rules (DRs) (Braun et al., 2001; Đurasević & Jakobović, 2018b; Maheswaran, Ali, Siegel, Hensgen, & Freund, 1999). DRs are simple heuristics that build the schedule iteratively by deciding which scheduling decision should be made next, i.e., which job should be scheduled on which machine. Because of this, they have limited visibility into the problem, which affects their performance. They can generate the schedule extremely fast, which makes them ideal for dynamic scheduling problems. On the other hand, metaheuristic algorithms search as much of the solution space as possible by starting with complete solutions and iteratively improving them by introducing changes (Hart, Ross, & Corne, 2005). Such methods can achieve extremely good results because they do not construct the schedule greedily, but rather try to improve the existing ones.

Various metaheuristics have been applied to the problem of scheduling on parallel unrelated machines. Researchers mainly used genetic algorithms (GAs) (Holland, 1992), tabu search (TS) (Glover, 1990), simulated annealing (SA) (Kirkpatrick, Gelatt, & Vecchi, 1983), ant colony optimisation (ACO), variable neighbourhood search (VNS), variable neighbourhood descent (VND) (Mladenović & Hansen, 1997), greedy randomised adaptive search procedure (GRASP) (Feo & Resende, 1995), and others. Among such a large number of methods, it is difficult to choose the one that is most appropriate for the problem under consideration. Moreover, the above algorithms are often combined into sophisticated methods that are difficult to understand and reproduce. In many cases, novel metaheuristics are also proposed to deal with scheduling problems. However, such methods usually offer limited novelty (Sörensen, 2015) or the problem in question could have been solved more effectively with a simpler method that yields competitive results (Molnar, Jakobović, & Pavelić, 2016).

Local search (LS) operators are popular methods for searching in the neighbourhood of the current solution. They have been used in various optimisation problems, including unrelated machines. Due to their increasing popularity, a wide range of LS operators have been proposed in the literature, along with metaheuristics that define how they should be applied. In addition, hybrid algorithms are proposed to further improve performance, which increases their complexity. This inevitably leads to the situation where it is difficult to choose the right LS operators or to determine which strategies should be used to apply them to a problem. However, instead of developing new LS operators or hybrid methods, one should consider if with already existing operators it is possible to construct simple methods that produce high-quality results.

This paper investigates whether various LS operators can be combined into simple but efficient methods for solving the problem of scheduling parallel unrelated machines. For this purpose, different approaches from the literature were analysed, based on which the most important parts of LS methods were identified, such as initial solution generation, perturbation and LS operators. Based on these observations, two simple iterative LS methods were defined. To provide an exhaustive analysis of the proposed methods, several LS operators were selected from the literature and used as building blocks for the proposed methods. The studied methods were compared with several metaheuristics and the results showed that the iterative LS procedures easily outperform all of the methods. These results indicate that simple LS based methods are as powerful as the more complex metaheuristics. This conclusion is important because it shows that it is more beneficial to focus on constructing simple LS based methods than to apply metaheuristics that may not be appropriate for the problem. The contributions of this work can be summarised as follows:

  • 1.

    Definition of two simple iterative LS methods and identification of the key design decisions in their development.

  • 2.

    Analyse the effectiveness of the various LS operators proposed in the literature.

  • 3.

    Investigation of the influence of different design decisions (initial solution generation, solution acceptance criterion, application of path relinking) on the performance of the simple iterative LS methods.

  • 4.

    Analysis of the performance of the simple iterative local search methods compared to several previously applied metaheuristics.

The paper is organised as follows. The literature review of heuristic methods applied to the unrelated machines environment problem is given in Section 2. Section 3 contains the description of the unrelated machines scheduling environment. The overview of the methods applied to solve the above problem is given in Section 4. Section 5 describes the experimental setup and outlines the obtained results. Finally, Section 6 concludes the paper and provides guidelines for future research.

Section snippets

Literature overview

Until now, a large number of studies have addressed the application of metaheuristics to the unrelated machines scheduling problem. Glass, Potts, and Shade (1994) applied GA, SA, and TS to optimise the makespan criterion. SA and TS used two neighbourhood operators, job reassignment and job interchange. The reassignment neighbourhood moved an job from one machine to another, while the interchange neighbourhood swapped two jobs on different machines. The results show that all three algorithms

Problem definition

The scheduling problem considered in this paper can be classified as R|rj|Twt (Pinedo, 2012), which means that the considered environment is the parallel unrelated machines environment with job release times and the total weighted tardiness criterion is optimised. A single problem instance in this environment consists of n jobs, where each job must be scheduled on one of the m available machines. It is assumed that the number of jobs and machines is finite. The subscript j refers to jobs, while

Iterated local search procedures

Over the years, several LS based methods have been proposed, such as GRASP (Feo & Resende, 1995), VNS (Hansen and Mladenovic, 2001, Mladenović and Hansen, 1997), and VND (Hansen, Mladenovic, Todosijević, & Hanafi, 2016). These methods work in similar ways, with the differences lying in details such as whether or not one or more LS operators are used, or whether or not the LS operators are applied in a deterministic manner. None of these procedures are explicitly used in this work, instead a

Experimental setup

To test the considered algorithms, a set of problem instances was generated using procedures from the literature (Kim et al., 2003, Lee et al., 2013, Lin et al., 2013, Vlašić et al., 2020). The generated set consists of 60 problem instances with different properties. The instances were generated with different combinations of the number of jobs and machines, with the number of jobs set to 12, 25, 50, and 100 and the number of machines set to 3, 6, and 10. Job processing times were generated

Conclusion

This paper investigated the performance of various metaheuristics for the parallel unrelated machines scheduling problem. The goal was to find out whether it is necessary to use complicated metaheuristics for the above problem, or whether it is possible to obtain satisfactory results using simpler methods. The experiments show that simple methods based on LS operators are more powerful than some complex or hybrid methods. Metaheuristics such as GA, ACO, and TS were not able to give as good

CRediT authorship contribution statement

Lucija Ulaga: Conceptualization, Software, Validation, Investigation, Visualization. Marko Đurasević: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Writing – original draft, Writing – review & editing, Visualization, Supervision. Domagoj Jakobović: Conceptualization, Validation, Writing – original draft, Writing – review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Funding

This work has been supported in part by Croatian Science Foundation, Republic of Croatia under the project IP-2019-04-4333.

References (67)

  • GlassC. et al.

    Unrelated parallel machine scheduling using local search

    Mathematical and Computer Modelling

    (1994)
  • HansenJ.V.

    Genetic search methods in air traffic control

    Computers & Operations Research

    (2004)
  • HansenP. et al.

    Variable neighborhood search: Principles and applications

    European Journal of Operational Research

    (2001)
  • JooC.M. et al.

    Rule-based meta-heuristics for integrated scheduling of unrelated parallel machines, batches, and heterogeneous delivery trucks

    Applied Soft Computing

    (2017)
  • KimD.-W. et al.

    Unrelated parallel machine scheduling with setup times using simulated annealing

    Robotics and Computer-Integrated Manufacturing

    (2002)
  • KimD.-W. et al.

    Unrelated parallel machine scheduling with setup times and a total weighted tardiness objective

    Robotics and Computer-Integrated Manufacturing

    (2003)
  • LiaoT. et al.

    A comparison of five hybrid metaheuristic algorithms for unrelated parallel-machine scheduling and inbound trucks sequencing in multi-door cross docking systems

    Applied Soft Computing

    (2014)
  • LogendranR. et al.

    Scheduling unrelated parallel machines with sequence-dependent setups

    Computers & Operations Research

    (2007)
  • MaheswaranM. et al.

    Dynamic mapping of a class of independent tasks onto heterogeneous computing systems

    Journal of Parallel and Distributed Computing

    (1999)
  • MladenovićN. et al.

    Variable neighborhood search

    Computers & Operations Research

    (1997)
  • RambodM. et al.

    Robust meta-heuristics implementation for unrelated parallel machines scheduling problem with rework processes and machine eligibility restrictions

    Computers & Industrial Engineering

    (2014)
  • RuizR. et al.

    A simple and effective iterated greedy algorithm for the permutation flowshop scheduling problem

    European Journal of Operational Research

    (2007)
  • UnluY. et al.

    Evaluation of mixed integer programming formulations for non-preemptive parallel machine scheduling problems

    Computers & Industrial Engineering

    (2010)
  • ValladaE. et al.

    A genetic algorithm for the unrelated parallel machine scheduling problem with sequence dependent setup times

    European Journal of Operational Research

    (2011)
  • VlašićI. et al.

    Improving genetic algorithm performance by population initialisation with dispatching rules

    Computers & Industrial Engineering

    (2019)
  • VlašićI. et al.

    A comparative study of solution representations for the unrelated machines environment

    Computers & Operations Research

    (2020)
  • AfzaliradM. et al.

    Design of high-performing hybrid meta-heuristics for unrelated parallel machine scheduling with machine eligibility and precedence constraints

    Engineering Optimization

    (2016)
  • Al-HarkanI. et al.

    Optimize unrelated parallel machines scheduling problems with multiple limited additional resources, sequence-dependent setup times and release date constraints

    IEEE Access

    (2019)
  • ArıkO.

    Comparisons of metaheuristic algorithms for unrelated parallel machine weighted earliness/tardiness scheduling problems

    Evolutionary Intelligence

    (2019)
  • ArnaoutJ.-P. et al.

    A two-stage ant colony optimization algorithm to minimize the makespan on unrelated parallel machines with sequence-dependent setup times

    Journal of Intelligent Manufacturing

    (2009)
  • BeanJ.C.

    Genetic algorithms and random keys for sequencing and optimization

    ORSA Journal on Computing

    (1994)
  • BurkeE. et al.

    The state of the art of nurse rostering

    Journal of Scheduling

    (2004)
  • ChengV. et al.

    Air traffic control using genetic search techniques

  • Cited by (9)

    • A variable neighborhood search algorithm for airport ferry vehicle scheduling problem

      2023, Transportation Research Part C: Emerging Technologies
    • Does size matter? On the influence of ensemble size on constructing ensembles of dispatching rules

      2023, GECCO 2023 Companion - Proceedings of the 2023 Genetic and Evolutionary Computation Conference Companion
    View all citing articles on Scopus
    View full text