A multi-start variable neighbourhood descent algorithm for hybrid flowshop rescheduling

https://doi.org/10.1016/j.swevo.2019.01.002Get rights and content

Highlights

  • Hybrid flowshop rescheduling considering three types of dynamic events is studied.

  • Approach for deciding lower and upper bounds of optimization objectives is presented.

  • A Multi-Start VND (MSVND) algorithm with a new hybrid decoding is proposed.

  • Intensification and diversification strategies are designed to enhance search power.

  • The effectiveness of the MSVND algorithm is demonstrated by extensive experiments.

Abstract

Hybrid flowshop (HFS) rescheduling has important applications in modern industry. Much of the existing research on HFS rescheduling only consider one type of dynamic event. However, realistic production systems often encounter several types of dynamic events. In this paper, HFS rescheduling considering simultaneously three types of dynamic events (i.e. machine breakdown, new job arrival and job release variation) is studied. The mathematical model of minimizing makespan and system instability is established. The approaches for calculating lower and upper bounds of the two optimization objectives are developed. A Multi-Start Variable Neighbourhood Descent (MSVND) algorithm is proposed for the HFS rescheduling. In the MSVND, a hybrid decoding is developed. To improve the intensification of the MSVND, a Fruit Fly Optimization (FFO)-based local search and an enhanced FFO-based local search are designed to improve the best solution found so far. Moreover, to enhance the diversification, a simulated annealing-like acceptance criterion is employed to determine whether the local optima can be accepted, and a restart strategy with perturbation is devised to guide the search to the so far unexplored area. Extensive experimental comparisons on 150 instances verify the effectiveness of the devised strategies. Further, a comprehensive comparison against seven highly efficient algorithms demonstrates the superiority of the MSVND.

Introduction

Hybrid Flowshop (HFS) scheduling is a complicated decision-making process of committing a number of jobs to a set of machines distributed in several stages [1,2]. HFS scheduling is a NP-hard combinatorial optimization problem and arises broadly in flexible manufacturing and process industry, such as electronics, paper, textile and Steelmaking-Continuous Casting (SCC) [1,3,4]. Efficient HFS schedules can improve production productivity greatly and make significant monetary savings. Therefore, the problem has attracted much attention in recent years [1,2,5].

A series of HFS scheduling methods have been reported in the literature, which can be categorized into three groups: mathematical programming methods, heuristics, and metaheuristics. Among the mathematical programming methods, branch and bound, Lagrangian relaxation have been extensively adopted [[6], [7], [8], [9], [10]]. As to heuristics, Gupta et al. [11] designed constructive heuristics for two-stage HFS scheduling where only the first stage contains identical machines. Kurza and Askin [12] proposed cyclic and insertion heuristics for HFS scheduling with Sequence Dependent Setup Times (SDST), while Dios et al. [5] provided a number of heuristics for HFS scheduling with missing operations.

With regard to the metaheuristics, early research mainly concentrated on HFS scheduling with traditional optimization objectives, i.e. minimizing makespan, total flowtime, or weighted completion time. Wang et al. [13] and Pan et al. [14] developed improved Estimation of Distribution Algorithm (EDA) and Artificial Bee Colony (ABC) respectively, where both of them designed novel decoding approaches. Pan and Dong [15] devised an improved Migrating Birds Optimization (MBO) algorithm for HFS. Tang and Wang [16] proposed a modified Particle Swarm Optimization algorithm for realistic HFS scheduling derived from SCC production. Li et al. [17] developed a hybrid Variable Neighbourhood Search (VNS), where a neighbourhood change strategy and EDA are integrated to enhance the diversification. In recent years, HFS scheduling with more realistic and complex characteristics has become increasingly active, since the characteristics such as setup times, due windows, and energy consumption have existed in the production process [[18], [19], [20]]. For example, to solve the HFS scheduling with SDST, Kurza and Askin [18], Ruiz and Maroto [19] presented effective Genetic Algorithms (GA), while Pan et al. [21] proposed three population-based and six trajectory-based methods, where the discrete ABC was demonstrated to be the best. Subsequently, for the HFS scheduling with due windows, Pan et al. [22] devised Iterated Greedy (IG) and Iterated Local Search. Recently, for the HFS scheduling with total/setup energy consumption, Lei et al. [23] and Li et al. [24] designed Teaching-Learning-Based Optimization (TLBO) and multi-objective evolutionary algorithm respectively.

HFS scheduling generally includes a series of assumptions, e.g. machines are always available for processing, jobs are available at time zero [1,25,26]. However, due to complicated production environments, realistic production systems often encounter a wide range of unexpected dynamic events, e.g. machine breakdown, new job arrival, and job release time variation, which would make the original schedule infeasible or ineffective, and have to be re-optimized [25,[27], [28], [29], [30]]. It is essential to respond to the dynamic events immediately and produce new schedules quickly, which corresponds to HFS rescheduling. HFS rescheduling is to rebuild a new optimal HFS schedule under production environments after dynamic events, which is also NP-hard [26]. Since the production environments after dynamic events are different from the original, static HFS scheduling methods may not be suitable for HFS rescheduling [25].

Unfortunately, however, compared with extensive studies on HFS scheduling, less attention has been paid to HFS rescheduling. Meanwhile, much of the existing work are based on metaheuristics. In the early studies, the traditional efficiency optimization objectives such as makespan are adopted, and machine breakdown was generally considered as the dynamic event. Mirabi et al. [30] devised a heuristic for the two-stage HFS rescheduling. Zandieh and Gholami [31] adopted an immune algorithm to solve HFS scheduling with SDST. Tang et al. [25] constructed a neural network model for HFS rescheduling with dynamic job arrivals, where average tardy time, the percentage of tardy jobs and average flow time were utilized as optimization objectives. Thereafter, Tang et al. [32] proposed a modified differential evolution algorithm to solve the SCC rescheduling. Tang et al. [33] developed a modified PSO for energy-efficient HFS rescheduling considering new job arrivals and machine breakdown, where makespan and energy consumption were used as optimization objectives. Moreover, differ from most of the existing research, Wang and Choi [34] and Wang et al. [35] devised decomposition-based approaches, the main idea is to decompose a rescheduling problem into several cluster scheduling problems first, and then solve them by different approaches. In the above researches, system instability is not considered usually, i.e. the difference in performance between new and original schedule. This is not consistent with the industrial reality. Recently, system instability has been given more attention and also been considered as one of the optimization objectives. Rahmani and Ramezanian [36] adopted VNS to address general HFS rescheduling considering new job arrival, where instability and total weighted tardiness were minimized. Li et al. [26] and Peng et al. [4] studied SCC rescheduling with machine breakdown, and devised a hybrid fruit fly optimization algorithm and an improved ABC, respectively.

Besides HFS rescheduling, other kinds of production rescheduling has also been studied in the literature, where Permutation Flow Shop (PFS) rescheduling and Flexible Job Shop (FJS) rescheduling have received more attention. For the PFS rescheduling, Katragjini et al. [29] devised a rescheduling framework under three dynamic events, i.e. machine breakdown, new job arrival, and job ready time variation. Lower and upper bounds for makespan and instability were provided, and an IG rescheduling algorithm was also developed. Later, in their extension work [37], rescheduling under five more dynamic events was further studied. Subsequently, Li et al. [38] proposed an effective TLBO when five different dynamic events were considered simultaneously, where an IG is integrated as local search. Liu et al. [39] provided four hybrid algorithms for rescheduling with the stochastic and dynamic events. Lu et al. [40] presented a hybrid multi-objective grey wolf optimization algorithm for rescheduling with controllable processing times., while Liu et al. [41] designed heuristics for rescheduling with new job arrival. For the PJS rescheduling considering new job arrival and fuzzy processing time, Gao et al. [42] developed an improved ABC, where probability-based solution generation operators were designed. Very recently, Gao et al. [43] developed a discrete Jaya algorithm for the FJS rescheduling with new job arrival, where five objective-oriented local search strategies were devised. It is worth noting that, although these PFS and PJS rescheduling methods cannot be applied for the HFS rescheduling directly, their main idea and schemes, e.g. objective-oriented local search, can be borrowed to devise the HFS rescheduling methods.

To summarize, there has been limited work dealing with HFS rescheduling in the literature. These researches mainly concentrate on two kinds of dynamic events: machine breakdown and new job arrival. Moreover, majority of the literature considered one and only one dynamic event. This is different from industrial standards since realistic production systems may suffer from several types of dynamic events. Therefore, HFS rescheduling simultaneously considering several dynamic events is essential to meet the demands of modern industry. However, to the best of our knowledge, the problem has seldom been investigated to this date. Thus, it is urgent to study the problem and devise effective HFS rescheduling methods.

Variable Neighbourhood Descent (VND) is an excellent trajectory-based metaheuristic for combinational optimization problems [[44], [45], [46], [47], [48]]. Due to the advantages of a few control parameters and powerful performance, the VND has been successfully utilized to solve a number of production scheduling problems, e.g. flowshop scheduling, block flowshop scheduling, HFS scheduling and distributed PFS scheduling [22,[49], [50], [51]]. However, to the best of our knowledge, there is little research on the application of the VND to deal with the HFS rescheduling.

In this paper, the HFS rescheduling problem considering simultaneously machine breakdown, new job arrival, and job release time variation is studied. Moreover, a Multi-Start Variable Neighbourhood Descent (MSVND) algorithm is developed to deal with the problem. More specifically, the main contributions are described as follows:

  • (1)

    The HFS rescheduling problem considering simultaneously three types of dynamic events is studied. The mathematical model with objectives of minimizing makespan and system instability is established.

  • (2)

    Approaches for determining lower and upper bounds of the optimization objectives are presented.

  • (3)

    A MSVND algorithm is devised to solve the HFS rescheduling problem.

  • (4)

    A hybrid decoding considering the problem characteristics is devised.

  • (5)

    Intensification and diversification strategies are devised to balance the exploitative and explorative tendencies of the search algorithm.

Extensive experimental comparisons demonstrate the efficiency of the specially-devised strategies. Additionally, comprehensive comparisons against seven state-of-the-art algorithms further verify the effectiveness of the MSVND. In these experiments, three combinations of objective weight coefficients and three termination conditions are tested.

The remainder of this paper is structured as follows. Section 2 introduces the HFS rescheduling problem in detail. Section 3 gives lower and upper bounds calculation approaches for the makespan and system instability. Section 4 proposes the devised MSVND for the HFS rescheduling. Section 5 provides the extensive computational experiments. Section 6 concludes the paper and unearths some avenues for future work.

Section snippets

Problem description

The HFS rescheduling problem studied in this paper consists of m stages (M = {1,2, …,m}). Every stage k ∈ M has nk identical parallel machines, where nk ≥ 2 for at least one stage. n jobs (J = {1,2, …,n}) are to be processed in the m stages. Each job j ∈ J must be processed sequentially from the 1st stage to the mth stage. Accordingly, each job j ∈ J has m operations: O1j,O2j, …,Omj. In each stage k ∈ M, each job j ∈ J must be processed on a machine i∈{1,2, …,nk} in this stage [14,15]. At time

Lower and upper bounds of optimization objectives

As the makespan and instability are in different units and have different orders of magnitude, normalization is utilized to scale their values into the range of [0,1] in this paper. Note that normalization has been used for the PFS rescheduling in Katragjini et al. [29] and Li et al. [38]. By applying the normalization, the objective function is obtained as follows:f=wf1N+(1w)f2Nf1N=f1f1Lf1Uf1Lf2N=f2f2Lf2Uf2Lwheref1Nandf2Ndenote the normalization values of the makespan and system

The proposed MSVND for HFS rescheduling

In this section, we discuss the proposed MSVND method in sufficient details. First, we introduce the basic VND. Subsequently, the encoding and hybrid decoding approach are elaborated. Then, we sequentially describe the initialization, three neighbourhood structures, neighbourhood changing strategy, intensification strategies, and diversification strategies. Finally, we provide a complete framework of the MSVND.

Experimental results

To demonstrate the effectiveness of the proposed MSVND, we compare it to seven effective algorithms in the literature, which are listed as follows:

  • (1)

    three high-performance flow shop rescheduling algorithms, i.e. discrete Teaching-Learning-Based Optimization algorithm (TLBO) in Li et al. [38], Iterated Greedy (IG) in Katragjini et al. [29], Iterated Greedy with Variable Neighbourhood Descent (IGVND), where the framework is same as that of the IG of Katragjini et al. [29], and the VND serves as the

Conclusions

In this paper, the HFS rescheduling problem considering simultaneously machine breakdown, new job arrival and job release time variation, is investigated. By hybridizing the VND with the specially-designed hybrid decoding, intensification and diversification strategies, an effective MSVND algorithm is developed for the problem. Seven effective algorithms have been adopted for comparison. Extensive computational comparisons on a total of 150 instances have confirmed the efficiency and

Acknowledgments

The authors sincerely thank the editor and anonymous reviewers for their valuable comments and helpful suggestions. This project is supported by National Natural Science Foundation of China (Grant No. 51705177, 51575212, 51825502, 51775216), Natural Science Foundation of Hubei Province (Grant No. 2018CFA078), and Program for HUST Academic Frontier Youth Team.

References (70)

  • Q.-K. Pan et al.

    An improved migrating birds optimisation for a hybrid flowshop scheduling with total flowtime minimization

    Inf. Sci.

    (2014)
  • J. Li et al.

    A hybrid variable neighborhood search for solving the hybrid flow shop scheduling problem

    Appl. Soft Comput.

    (2014)
  • M.E. Kurz et al.

    Scheduling flexible flow lines with sequence-dependent setup times

    Eur. J. Oper. Res.

    (2004)
  • R. Ruiz et al.

    A genetic algorithm for hybrid flowshops with sequence dependent setup times and machine eligibility

    Eur. J. Oper. Res.

    (2006)
  • Q.-K. Pan et al.

    Effective metaheuristics for scheduling a hybrid flowshop with sequence-dependent setup times

    Appl. Math. Comput.

    (2017)
  • Q.-K. Pan et al.

    Iterated search methods for earliness and tardiness minimization in hybrid flowshops with due windows

    Comput. Oper. Res.

    (2017)
  • J. Li et al.

    Efficient multi-objective optimization algorithm for hybrid flow shop scheduling problems with setup energy consumptions

    J. Clean. Prod.

    (2018)
  • D. Tang et al.

    Energy-efficient dynamic scheduling for a flexible flow shop using an improved particle swarm optimization

    Comput. Ind.

    (2016)
  • D. Rahmani et al.

    A stable reactive approach in dynamic flexible flow shop scheduling with unexpected disruptions: a case study

    Comput. Ind. Eng.

    (2016)
  • J. Li et al.

    A discrete teaching-learning-based optimisation algorithm for realistic flowshop rescheduling problems

    Eng. Appl. Artif. Intell.

    (2015)
  • C. Lu et al.

    A hybrid multi-objective grey wolf optimizer for dynamic scheduling in a real-world welding industry, Eng

    Appl. Artif. Intell.

    (2017)
  • K.Z. Gao et al.

    Artificial bee colony algorithm for scheduling and rescheduling fuzzy flexible job shop problem with new job insertion

    Knowl. Base Syst.

    (2016)
  • P. Chen et al.

    Iterated variable neighborhood descent algorithm for the capacitated vehicle routing problem

    Expert Syst. Appl.

    (2010)
  • Z. Zeng et al.

    Iterated tabu search and variable neighborhood descent for packing unequal circles into a circular container

    Eur. J. Oper. Res.

    (2016)
  • Z. Zeng et al.

    Adaptive tabu search and variable neighborhood descent for packing unequal circles into a square

    Appl. Soft Comput.

    (2018)
  • J. Gao et al.

    A hybrid genetic and variable neighborhood descent algorithm for flexible job shop scheduling problems

    Comput. Oper. Res.

    (2008)
  • B. Naderi et al.

    The distributed permutation flowshop scheduling problem

    Comput. Oper. Res.

    (2010)
  • M.F. Tasgetiren et al.

    Iterated greedy algorithms for the blocking flowshop scheduling problem with makespan criterion

    Comput. Oper. Res.

    (2017)
  • J. Li et al.

    Solving the large-scale hybrid flow shop scheduling problem with limited buffers by a hybrid artificial bee colony algorithm

    Inf. Sci.

    (2015)
  • F. Neri et al.

    Memetic algorithms and memetic computing optimization: a literature review

    Swarm Evol. Comput.

    (2012)
  • M. Lozano et al.

    Hybrid metaheuristics with evolutionary algorithms specializing in intensification and diversification: overview and progress report

    Comput. Oper. Res.

    (2010)
  • N. Lynn et al.

    Heterogeneous comprehensive learning particle swarm optimization with enhanced exploration and exploitation

    Swarm Evol. Comput.

    (2015)
  • J. Ding et al.

    Breakout dynasearch for the single-machine total weighted tardiness problem

    Comput. Ind. Eng.

    (2016)
  • B. Zhang et al.

    A hybrid variable neighborhood search algorithm for the hot rolling batch scheduling problem in compact strip production

    Comput. Ind. Eng.

    (2018)
  • W. Pan

    A new fruit fly optimization algorithm: taking the financial distress model as an example

    Knowl. Base Syst.

    (2012)
  • Cited by (74)

    • A systematic review of multi-objective hybrid flow shop scheduling

      2023, European Journal of Operational Research
    View all citing articles on Scopus
    View full text