Fixed set search application for minimizing the makespan on unrelated parallel machines with sequence-dependent setup times
Introduction
One of the most important practical problems encountered in manufacturing is the optimal use of existing machines. In many cases it is related to scheduling different types of jobs on one or multiple machines. This basic setting can be related to a wide range of operational problems. This type of problem has been formalized using the concept of job scheduling problems, commonly by the parallel machine scheduling problem (PMSP). Due to the use of this model for diverse real-world systems, many variations have been defined to address their specific properties. Some settings are based on relations between different machines, precedence constraints between jobs [1], [2], resource constraints [3], [4], maintenance [5], job incompatibilities [6], limited information [7], multi-objective [8], etc.
In this paper the focus is on the setting with unrelated parallel machines with machine-dependent and job sequence-dependent setup times. In it, the time a job needs to be processed is dependent on the machine and on the preceding job. This problem setting has been used in combination with a variety of objective functions like minimization of the makespan [9], [10], [11], total completion time [12], weighted tardiness of the jobs [13], and many others. Mixed integer programs (MIP) have been proposed for finding optimal solutions for the PMSP [14], [15]. An alternative, more efficient, MIP formulation has been presented in [16], for which several variations are explored in [17]. Branch-and-bound algorithms have also been used to find optimal solutions for this type of problem [18], [19].
In this paper, the focus is on the PMSP with unrelated parallel machines and sequence-dependent times with the objective of minimizing the makespan for which the common notation is . This variation of the PMSP is an NP-hard problem, since even the simpler special case of identical parallel machine scheduling, with the objective of minimizing the makespan is NP-hard even in the case of only two machines [20], [21]. Due to this fact, a wide range of heuristic and metaheuristic methods have been designed to find near-optimal solutions. In [14], a basic heuristic approach has been proposed. Some of the earlier works on the application of metaheuristics to this problem are dedicated to tabu search [22] and simulated annealing [23], [24], [25]. Population-based metaheuristics like artificial bee colony [26] and genetic algorithms [15], [27] have been applied with great success to the PMSP. In case of the PMSP the most efficient approach has been the combined use of MIP and heuristics in a matheuristic setting [17]. Although this approach manages to outperform standard metaheuristics it necessitates an extensive and complex implementation.
The use of metaheuristics based on the greedy algorithm in combination with local searches has proven to be most promising for the PMSP in the sense of a good balance between the complexity of implementation and the quality of found solutions. In [9], an interesting method named Meta-RS has been presented. In practice this method is highly similar to the Greedy Randomized Adaptive Search Procedure (GRASP) [28]. An ant colony optimization (ACO) algorithm has been proposed in [29] and later improved in [10] which significantly outperforms the Meta-RS. The same authors applied the Worm Optimization (WO), another metaheuristic based on the constructive greedy algorithm, which improves the learning mechanism of ACO [11]. The WO produced further improvements in the quality of found solutions.
Recently, a new metaheuristic based on the repetitive use of greedy algorithms, the Fixed Set Search (FSS) has been introduced and successfully applied to the Traveling Salesman Problem (TSP) [30], the Minimum Weighted Vertex Cover Problem (MWVCP) [31] and the Power Dominating Set Problem [32]. The novelty in the design of the FSS, compared to ACO and similar methods, is that it assumes that a local search plus further simple modifications are used. In practice, it uses a simple approach to add a learning mechanism to the GRASP metaheuristic. The idea of the FSS is to exploit the fact that some elements frequently appear in high quality solutions. In this paper the FSS is applied to the PMSP. The importance of this application is in extending the number of problem families to which the FSS can successfully be applied. The FSS has previously been applied to routing problems [30] and to covering problems with limited [31] as well as unlimited distance of effect [32]. In this way it has been confirmed that the FSS can be applied to very different types of combinatorial optimization problems.
The performed computational experiments show that the FSS manages to outperform state-of-the-art methods in the quality of found solutions. The FSS achieves this at a similar computational cost. Current research on metaheuristics based on the greedy algorithm lacks the evaluation in view of optimal solutions and MIPs in general. Because of this we have conducted additional computational experiments to evaluate when the use of such metaheuristics is justified.
The paper is organized as follows. In the next section the problem formulation is given in a graph form suitable for use in the FSS. In Section 3 the greedy algorithm used as a base for the FSS is presented. In the following section, the used local search is presented and how the greedy algorithm is extended to the GRASP metaheuristic. In Section 5 details of the FSS and its application to the PMSP are given. The subsequent section is dedicated to the discussion of the performed computational experiments. Section 7 concludes.
Section snippets
Problem formulation
The PMSP has the following setting. There is a set of jobs which are scheduled on a set of machines . A schedule of a machine is a sequence (array) of jobs . A schedule on a set of machines is a set of machine schedules . Each job is only processed once on a single machine. Each job has an associated processing time related to a parallel machine . For each job , there is a setup time if it is processed immediately after job
Greedy algorithm
In this section the used greedy algorithm for the PMSP is presented. The general approach in developing a greedy algorithm for the PMSP is to iteratively extend a partial solution . Here a partial solution is a set of sequences of jobs performed on each machine. It is extended by adding a job to a schedule on a machine . Adding a job to such a schedule can be understood as processing it after some job (), where can also be the dummy job but not . The job , machine and
GRASP
In this section the extension of the proposed greedy algorithm to the GRASP metaheuristic is presented. To achieve this, the randomization greedy algorithm (RGA) and a local search need to be defined. The randomization is done in the standard way by using a restricted candidate list (RCL). In practice this means that instead of selecting the candidate for expansion having the maximal value of the heuristic function , one of the best candidates is selected randomly. We wish to point out that
Fixed set search
In this section, we give details on how the FSS metaheuristic is applied to the PMSP. The FSS adds a learning mechanism to the GRASP. The basic idea is to recognize elements that frequently occur in high quality solutions and generate new solutions that contain such elements. This idea has previously been successfully applied to the TSP [30] and the MWVCP [31]. In the following text the general concepts appearing in the FSS are explained and how they are applied to the PMSP.
The FSS exploits the
Results
In this section, the results of the conducted computational experiments are presented. The first objective was to make a comparison between the FSS and current state-of-the-art population-based metaheuristics for solving the PMSP with the objective of minimizing the makespan. The second objective was to evaluate the FSS against optimal solutions and MIP programs, with the intention of evaluating when the use of the FSS is most suitable.
The used parameters for the FSS are the following:
Conclusion
The problem of minimizing the makespan on unrelated parallel machines with sequence-dependent setup times has been solved using the FSS. To achieve this the PMSP has been presented in a graph and the standardly used greedy algorithm has been adapted to it. Further, the greedy algorithm has been extended to the GRASP metaheuristic by adding randomization to it and combining it with a local search. The GRASP algorithm has been used as a basis for the FSS. In defining the FSS, the randomized
CRediT authorship contribution statement
Raka Jovanovic: Conceptualization, Methodology, Software, Investigation, Writing - review & editing, Writing - original draft. Stefan Voß: Conceptualization, Methodology, Investigation, 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.
References (37)
- et al.
List scheduling in a parallel machine environment with precedence constraints and setup times
Oper. Res. Lett.
(2001) - et al.
A divide and merge heuristic for the multiprocessor scheduling problem with sequence dependent setup times
European J. Oper. Res.
(2001) - et al.
Models and matheuristics for the unrelated parallel machine scheduling problem with additional resources
European J. Oper. Res.
(2017) - et al.
Algorithms for the unrelated parallel machine scheduling problem with a resource constraint
European J. Oper. Res.
(2018) - et al.
Including preventive maintenance activities in an unrelated parallel machine environment with dependent setup times
Comput. Ind. Eng.
(2018) - et al.
A matheuristic framework for batch machine scheduling problems with incompatible job families and regular sum objective
Appl. Soft Comput.
(2018) - et al.
Ant colony optimization algorithm for scheduling jobs with fuzzy processing time on parallel batch machines with different capacities
Appl. Soft Comput.
(2019) - et al.
A realistic variant of bi-objective unrelated parallel machine scheduling problem: NSGA-II and MOACO approaches
Appl. Soft Comput.
(2017) - et al.
A tabu search heuristic for the multiprocessor scheduling problem with sequence dependent setup times
Int. J. Prod. Econ.
(1996) - et al.
Scheduling jobs on parallel machines with sequence-dependent setup times
European J. Oper. Res.
(1997)
A genetic algorithm for the unrelated parallel machine scheduling problem with sequence dependent setup times
European J. Oper. Res.
Reformulations and an exact algorithm for unrelated parallel machine scheduling problems with setup times
Comput. Oper. Res.
Scheduling unrelated parallel machines to minimize total weighted tardiness
Comput. Oper. Res.
Scheduling jobs with release dates and tails on two unrelated parallel machines to minimize the makespan
European J. Oper. Res.
Unrelated parallel machine scheduling with setup times using simulated annealing
Robot. Comput.-Integr. Manuf.
Unrelated parallel machine scheduling with setup times and a total weighted tardiness objective
Robot. Comput.-Integr. Manuf.
ABC-based manufacturing scheduling for unrelated parallel machines with machine-dependent and job sequence-dependent setup times
Comput. Oper. Res.
Integrating dominance properties with genetic algorithms for parallel machine scheduling problems with setup times
Appl. Soft Comput.
Cited by (22)
Learning-based multi-start iterated local search for the profit maximization set covering problem
2023, Information SciencesFixed set search applied to the clique partitioning problem
2023, European Journal of Operational ResearchJoint Planning of Production and Inspection of Parallel Machines with Two-phase of Failure
2022, Reliability Engineering and System SafetyCitation Excerpt :Different measures are used for parallel machines production environments, such as a fraction of on-time delivery and makespan. Considering the makespan as the measure for singular or parallel machines production environment is a proper way to minimize the jobs’ total completion time and on-time delivery of the jobs (i.e., [7,18,24,25]). Since in this paper we consider an unrelated identical parallel machines production environment, minimizing the makespan is selected as the objective.
Initialization of metaheuristics: comprehensive review, critical analysis, and research directions
2023, International Transactions in Operational Research