Q-learning-based Hierarchical Cooperative Local Search for Steelmaking-continuous Casting Scheduling Problem

Bin Qian; Jian-Bo Yang; Rong Hu; Yang Lv

REVIEW 6 major objections 6 minor 37 references

Q-learning-based Hierarchical Cooperative Local Search for Steelmaking-continuous Casting Scheduling Problem

T0 review · 6 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper claims that a Q-learning local search, rewarded for improving a new Coupling Measure between charge and cast sequences, finds better steelmaking-continuous casting schedules in a fraction of the time of existing metaheuristics.

desk verdict Useful scheduling heuristic with a genuinely new coupling-measure reward, but a broken NP-hardness proof and hidden baseline tuning mean it needs serious refereeing, not blind trust. read the letter →

arxiv 2506.08608 v1 pith:2TIF5XUR submitted 2025-06-10 eess.SY cs.SY

Yang Lv , Rong Hu , Bin Qian , Jian-Bo Yang This is my paper

classification eess.SYcs.SY MSC 90B3590C5968T20

keywords steelmaking-continuouscastingschedulingQ-learningcouplingmeasurelocalsearchhierarchicaloptimizationno-waithybridflowshopreinforcementlearningproduction

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Steelmaking-continuous casting scheduling requires ordering charges through steelmaking and refining, then grouping them into casts on continuous-casting machines, with no-wait constraints and costly reheating if timing slips. This paper claims that high-quality schedules occupy a small subspace in which the charge order aligns with the order implied by the cast sequence, and that this alignment can be measured and used as a learning signal. It builds HierC_Q, a Q-learning-based hierarchical cooperative local search: three Q-learning local searchers refine the charge subproblem, the cast subproblem, and their joint state, while a disturb-to-renovate tier perturbs and reconstructs solutions to escape local optima. The reward function gives credit for improving the proposed Coupling Measure even when the objective does not move, steering search toward that promising subspace. On 20 randomly generated instances, HierC_Q reports the lowest average relative percentage deviation among eleven local-search frameworks and nine state-of-the-art algorithms under equal runtime, and nearly the best results among all methods at one-third the runtime.

What carries the argument

The load-bearing mechanism is the Coupling Measure (CM) of Eq. (15), built from a Gaussian fuzzy relation matrix: for a schedule $\pi=(u,v)$, construct the virtual charge sequence by concatenating charges cast by cast; set membership $r(l_{e-1}+k,j)=\exp\{-[\mathrm{Pos}(l_{e-1}+k)-j]^2/(2\sigma^2)\}$; then $\mathrm{CM}(\pi)=\frac{1}{N}\sum_{i=1}^N r(i,i)$. High CM means each charge sits near its cast-implied position. The reward function Eq. (22) couples this proxy to Q-learning: actions that reduce fitness but increase CM still earn a small positive reward, which is the bet that tight coupling leads to the true optimum. Around this mechanism, the search alternates three Q-tables (charge, cast, joint) acting on eight charge neighborhood operators and three cast operators; Theorem 1 prunes invalid charge moves by checking whether an operator inverts the within-cast processing priority, and Theorem 2 prunes cast-move evaluations using the key cast on the critical machine.

What would settle it

Compute the rank correlation between $\Delta \mathrm{CM}$ and $\Delta f$ over many random neighborhood moves on a fixed SCCSP instance. If the correlation is not positive, or if removing the $+0.2$ CM-only reward branch from Eq. (22) does not degrade HierC_Q's ARPD, then the coupling hypothesis is not carrying the reported gain.

Watch

Extended reading notes

Core claim

The central claim, in the paper's own terms, is that SCCSP decomposes into two coupled subproblems and that a schedule's quality is strongly tied to how tightly the charge subsequence $u$ is coupled to the cast subsequence $v$. The paper formalizes this with a Coupling Measure: from $v$ it builds a virtual charge sequence $u_{\mathrm{virtual}}$ by listing each cast's charges in processing-priority order; a Gaussian membership function assigns each charge a fuzzy position, producing a fuzzy relation matrix $R_{\mathrm{charge}}$; and $\mathrm{CM}(\pi)=\frac{1}{N}\sum_i r(i,i)$ measures how close the actual charge order is to the virtual one. The reward function (Eq. 22) then gives $+1.5$ when both fitness and CM improve, $+1$ when fitness improves alone, $+0.2$ when only CM improves, and zero otherwise. Three cooperative Q-learning frameworks (Charge_QLSF, Cast_QLSF, and a synergy SQLSF) select among eight charge-specific neighborhood operators and three cast operators; a validity theorem filters charge moves that break within-cast priority, and a critical-machine theorem accelerates cast-move evaluation. The paper's reported results are that HierC_Q achieves lower ARPD than all compared algorithms under equal runtime on every instance, and that at runtime factor $\lambda=200$ it beats nearly all competitors even when they are allowed $\lambda=400$.

Load-bearing premise

The load-bearing premise is that moving charges closer to the order implied by their casts (higher Coupling Measure) reliably moves schedules closer to the true optimum, so rewarding CM gains even without immediate fitness improvement guides search in the right direction.

Editorial extensions

If this is right

Under equal runtime limits, HierC_Q reports lower ARPD than all eleven local-search frameworks and all nine state-of-the-art algorithms on every one of the 20 generated instances.
At the shortest runtime factor ($\lambda=200$), HierC_Q still outperforms nearly all competing algorithms that are given twice the time ($\lambda=400$), implying roughly a threefold runtime advantage at comparable quality.
Problem-dependent neighborhood operators for charges and casts beat traditional Swap, Insert, and Exchange variants of the same frameworks, so structure-aware moves carry part of the gain.
The disturb-to-renovate renewal strategy prevents stagnation, and the CM-based reward is credited with focusing search on the tightly coupled subspace where good schedules live.
The framework is claimed to extend beyond SCCSP to other decomposable scheduling problems such as the assembly job shop scheduling problem and integrated scheduling.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Editorial inference: if the coupling hypothesis holds, CM could be computed incrementally after each local move, turning the current $O(N)$ reward evaluation into a faster update and widening the runtime advantage.
Editorial inference: the same reward design could transfer to other two-level scheduling problems where a secondary sequence is induced by a primary sequence; a natural test is whether CM-based Q-learning beats objective-only rewards there.
Editorial inference: the reported runtime gain suggests the method's value lies less in asymptotic complexity and more in guiding search effort to valid, promising moves; one could verify this by profiling the fraction of time spent on valid versus invalid moves.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Best to know: HierC_Q is a serious piece of engineering for steelmaking-continuous casting scheduling. The genuinely new pieces are the coupling-measure reward and the three-agent hierarchical Q-learning local search with parameterized neighborhood operators. The experimental campaign is large: 20 instance families, 30 runs each, three time budgets, 20 comparators. The average ARPD and SD numbers favor HierC_Q almost everywhere, and the claim that it beats most competitors at half the runtime is credible if the baselines are fair.

What the paper does well: Q-learning operator selection is not new (refs 2, 28, 29 are honest about that), but the CM-based reward in Eq. (22) that rewards a fuzzy coupling gain even without immediate fitness improvement is a clever idea, and the decomposition into charge/cast/synergy QLSFs with validity and speed-up evaluations is a reasonable use of problem structure. The paper gives credit for prior Q-learning work, so the novelty claim is limited to the coupling measure and the hierarchy, which is appropriate.

Soft spots, in order of severity. First, the NP-hardness proof is broken. The text says it reduces from 1||Cmax, which is polynomial, so it cannot establish strong NP-hardness. Either the reduction is misdescribed or the proof is wrong. This does not invalidate the algorithm, but it is a correctness claim the paper makes as a contribution. Second, the key evidence is partly outside the arXiv record: baseline parameter settings, Tukey HSD plots, parameter calibration, and ablations are all in the supplementary or GitHub without a commit hash. The stress-test concern about under-tuned baselines is not fully resolved; for the state-of-the-art comparison in Table III, we see only the authors' reimplementations and no evidence of published settings. For Table II, the comparison is explicitly against LSFs equipped with the authors' PDNOs, which is a fair ablation of the Q-learning selection mechanism, not a test of the baselines as published. Third, the CM hypothesis is validated only on self-generated instances with ARPD anchored to the best solution found in the comparison set; no external benchmark anchors absolute quality. Minor issues: Algorithm 3 uses the wrong episode constant, Algorithm 6 calls SLF instead of SQLSF, and the 'first local search approach to SCCSP' claim contradicts the paper's own comparison set.

Who this is for: people working on steelmaking scheduling and anyone combining Q-learning with local search for coupled decomposable problems. The central mechanism is plausible and the engineering is thorough. It deserves a serious referee, but the referee should demand the supplementary data, a corrected complexity proof, and sensitivity analysis on baseline parameters.

Referee Report

6 major / 6 minor

Summary. The paper proposes HierC_Q, a Q-learning-based hierarchical cooperative local search framework for the steelmaking-continuous casting scheduling problem (SCCSP). The method combines a coupling measure (CM) based reward, three cooperative Q-learning-based local search frameworks for the charge, cast, and joint subproblems, and a perturb-and-reconstruct diversification tier. The authors claim that HierC_Q outperforms eleven local search frameworks and nine state-of-the-art algorithms under equal or lower runtime, based on extensive experiments on 20 self-generated instance classes with 30 runs per instance and multiple runtime budgets.

Significance. If the performance claims are validated, the CM-guided reward and the hierarchical QLSF architecture would be a useful addition to the scheduling metaheuristic literature, particularly for decomposable problems with coupled subproblems. The experimental effort is substantial: 20 instance classes, 30 independent runs, three runtime factors, and 20 comparison algorithms. However, the current evidence is weakened by the reliance on self-generated instances, an ARPD definition anchored to the best solution found within the comparison set, incomplete reporting of baseline parameter settings, and an internal contradiction in the claimed universal superiority.

major comments (6)

[Section V-B, Table II] The text states that 'HierC_Q consistently achieves lower ARPD values than all other algorithms under the same λ, across all instances,' but Table II contradicts this. For instance 3×30, HHGAvar achieves ARPD 0.187 versus HierC_Q's 0.496, and for 4×15, ALNSvar and ALNS achieve 0.089 and 0.084 versus HierC_Q's 0.367. This directly undermines the universal-dominance claim and needs to be corrected or substantially qualified.
[Section V-A, Eq. (28)] ARPD is defined relative to the best objective value found by the compared algorithms on the same self-generated instances. This measures only relative ranking within the comparison set, not absolute solution quality. Since no external optimum, lower bound, or independently published benchmark is used, the claim of 'better schedule quality' at one-third runtime is not established. The authors should add validation on small instances against exact methods or against lower bounds, or at least clearly state that the results are only relative rankings.
[Section V-B] The five local search baselines (VNS, ILS, GD, ALNS, HHGA) are 'equipped with the PDNOs and perturbation operators introduced in this study to ensure a fair and thorough comparison.' This makes the comparison primarily a test of the Q-learning-based operator-selection mechanism, not a comparison against the published local search frameworks as used in the literature. The parameter configurations for the state-of-the-art group are deferred to Part 8 of the supplementary material, which is missing from the arXiv record. Without these settings and a sensitivity analysis, the possibility that the baselines are under-tuned cannot be excluded.
[Section III-B and Section IV-B, Eq. (22)] The CM-based reward function includes a branch that rewards coupling-measure improvement even when the fitness does not improve (the 0.2 branch). The paper asserts that 'optimal solutions typically reside in tightly coupled subspaces,' but this is not empirically tested. A correlation or ablation study showing that CM improvements are associated with objective improvements, together with sensitivity analysis for σ and the reward magnitudes, is needed to justify the central design choice.
[Section IV-B, Eq. (27)] The Q-table update rule is Q(s_i,a_j) = (1-α)Q(s_i,a_j) + α r(s_i,a_j). This is a reward averaging scheme without the next-state max term or discount factor of standard Q-learning. As written, the method is not Q-learning but an adaptive operator-selection mechanism with a tabular reward memory. This distinction matters for the paper's central claim of a 'Q-learning-based' framework and should be clarified or corrected.
[Section IV-C, Theorems 1 and 2] The validity and speed-up theorems are load-bearing for the claimed computational efficiency, but their proofs are relegated to Part 4 of the supplementary material, which is not included in the arXiv submission. I could not verify these results. The authors should include the proofs or the full supplementary material with the submission.

minor comments (6)

[Algorithms 1 and 3] Algorithm 1 line 5 and Algorithm 3 line 6 refer to Eq. (26) for the ε schedule, but the adaptive ε formula is Eq. (25). Algorithm 3's while condition uses Epcast instead of Epjoint.
[Table II caption] The caption says 'UNDER Λ=200 AND 400,' but the table as presented lists only λ=200 values. The caption and table should be reconciled.
[Table III] The column headers for ARPD and SD are not clearly separated; the table would be much easier to read with explicit 'ARPD' and 'SD' subheaders for each algorithm.
[Section IV-C, Theorem 2] The statement of Theorem 2 contains garbled notation and unmatched parentheses, making it difficult to parse even before consulting the supplementary proof.
[Section V-A] The GitHub repository URL is written with spaces ('https:// github. com /ly726564418 /TCYBE .git') and should be formatted as a single clickable link.
[Section VI] The claim that this is 'the first work to apply a local search-based approach to the SCCSP' is overstated given the many local-search-based metaheuristics cited and compared in the paper; please soften or clarify the scope.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: HierC_Q's design uses a heuristic CM-based reward and an online Q-learning operator selector, and its reported advantage is an empirical comparison rather than a quantity forced by construction.

full rationale

The paper's central claim is algorithmic: a Q-learning-based hierarchical local search with a coupling-measure reward is asserted to outperform other local-search frameworks and published SCCSP algorithms on generated instances. I checked for definitional reductions and for predictions that are fitted inputs renamed. The CM-based reward (Eq. 22) rewards simultaneous improvement of fitness and CM, or CM improvement alone; this is a heuristic proxy for solution quality, but CM is not derived from the objective function and the objective is not defined as CM. Thus there is no self-definitional collapse. The Q-learning tables are initialized to zero and updated online on each instance; no model is pre-fitted to the test instances and then reported as a prediction, so no fitted-input-called-prediction pattern applies. The ARPD metric (Eq. 28) is relative to the best solution found by all compared algorithms; however, HierC_Q does not always find that best solution (its average ARPD in Table III is 0.687, not zero), so its lower ARPD is not guaranteed by construction. The decision to equip all compared LSFs with the paper's own PDNOs and perturbation operators, and the deferral of baseline parameter settings and ablation studies to supplementary material, are legitimate concerns about benchmarking fairness and verifiability, but they are not circularity. The only self-citations (e.g., refs. [17] and [26]) support general statements about neighborhood operators and SCCSP context and are not load-bearing for the paper's core result. The NP-hardness proof and several theorems are deferred to supplementary material, which is missing from the arXiv record; that is a completeness problem, not a circular-derivation problem. Overall, the paper's derivation chain does not reduce to its own inputs, and no specific equation or claim is shown to be equivalent to a fit or to a self-citation.

Assumptions & free parameters 8 free parameters · 5 assumptions · 3 invented entities

The central design rests on a coupling hypothesis that is asserted, not derived, and on a batch of hand-set learning parameters (sigma, reward magnitudes, episode lengths, patience). The model itself assumes a fixed charge route and predefined cast groupings. The only independent evidence the paper offers is its own experimental comparisons on self-generated instances.

free parameters (8)

sigma (Gaussian variance in CM membership function) = not stated in main text; tuned in supplementary
Controls the sharpness of positional relevance in Eq. (13); hand-tuned.
Reward magnitudes in Eq. (22) = 1.5, 1, 0.2, 0
Hand-set reward values for the four fitness/CM outcome cases; no sensitivity analysis in main text.
Q-learning rate alpha = not stated in main text; in supplementary
Linear weighting update rule in Eq. (27); hand-set.
Epsilon schedule endpoints epsilon_0 and epsilon_max = not stated in main text; in supplementary
Bound the adaptive exploration rate in Eq. (25); hand-set.
Patience parameter gamma (D2R trigger) = not stated in main text; in supplementary
Number of consecutive non-improving iterations before perturbation; hand-set.
Episode periods Ep_charge, Ep_cast, Ep_joint = not stated in main text; in supplementary
Loop bounds for the three QLSFs in Algorithms 1-3; hand-set.
PDNO distance thresholds = N/6 and N/2 boundaries
Partition swap and insert into small, medium, and large distance classes; chosen by hand.
Objective weights Psi_1 and Psi_2 = 10 and 1
Weighted sum of Cmax and average waiting time in Eq. (10); stated as reflecting industrial priorities, adjustable.

assumptions (5)

domain assumption All charges follow the identical route steelmaking, refining, continuous casting, with predefined cast groupings.
Section II; the entire permutation model is built on this route structure.
ad hoc to paper Tightly coupled charge-cast sequences contain the optimal solutions.
Section III-A; the load-bearing heuristic premise behind the CM reward; validated only in this paper's experiments.
ad hoc to paper The Gaussian function is the right membership model for positional relevance.
Eq. (13); the choice of sigma controls CM values but no sensitivity analysis appears in the main text.
standard math The base problem 1||Cmax is strongly NP-hard.
Section II; as rendered this is false, 1||Cmax is polynomially solvable, so the reduction as stated cannot establish strong NP-hardness.
ad hoc to paper The validity and speed-up theorems (Theorems 1 and 2) hold.
Section IV-C; proofs are deferred to supplementary material and cannot be verified from the main text.

invented entities (3)

Coupling Measure (CM)
purpose: A scalar score of charge-cast sequence alignment used to shape the Q-learning reward.
Defined in Section III-B, Eqs. (13)-(15); its only support is the paper's own experiments.
Virtual charge subsequence u_virtual
purpose: A reference sequence constructed from the cast subsequence, used to compute CM.
Eq. (11); internal analytic construct.
Fuzzy relation matrix R_charge
purpose: Membership matrix comparing the current charge sequence to the virtual one.
Eq. (14); internal analytic construct.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Q-learning-based Hierarchical Cooperative Local Search for Steelmaking-continuous Casting Scheduling Problem." pith.science (2026). https://pith.science/paper/2TIF5XUR

@misc{pith2026250608608,
  author       = {Pith},
  title        = {Pith review of: Q-learning-based Hierarchical Cooperative Local Search for Steelmaking-continuous Casting Scheduling Problem},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2TIF5XUR}},
  note         = {Machine review of arXiv:2506.08608}
}

read the original abstract

The steelmaking continuous casting scheduling problem (SCCSP) is a critical and complex challenge in modern steel production, requiring the coordinated assignment and sequencing of steel charges across multiple production stages. Efficient scheduling not only enhances productivity but also significantly reduces energy consumption. However, both traditional heuristics (e.g., two-stage local search) and recent metaheuristics often struggle to adapt to the dynamic characteristics of practical SCCSP instances. To address these limitations, this paper introduces a novel Q learning based hierarchical cooperative local search framework, termed HierC_Q, aimed at minimizing the weighted sum of the maximum completion time and the average waiting time in SCCSP. The core contributions of HierC_Q are twofold. First, considering the intrinsic coupling properties of the SCCSP, a dedicated reward function is proposed based on a novel coupling measure (CM), guiding the exploration process towards promising regions of the solution space. Second, a hierarchical architecture is devised, comprising two distinct tiers: the learn to improve (L2I) tier and the "disturb to renovate" (D2R) tier. The L2I tier performs deep exploitation within promising regions using two independent Q-learning-based local search frameworks (QLSFs) tailored for subproblems, along with a synergy QLSF designed for the main problem. To enhance the effectiveness of local search, a validity evaluation approach and a speed-up evaluation method are also intro-duced, grounded in a detailed study of the problem's structure. Meanwhile, the D2R tier incorporates a perturbation and construction based solution renewal strategy to mitigate the risk of premature convergence. The superiority and effectiveness of HierC_Q are demonstrated through extensive comparisons with eleven local search frameworks and nine state-of-the-art algorithms.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

37 extracted references · 37 canonical work pages

[1]

Production scheduling in dynamic real-time optimization with closed-loop prediction,

J. E. Remigio, and C. L. Swartz, “Production scheduling in dynamic real-time optimization with closed-loop prediction,” Journal of Process Control, vol. 89, pp. 95-107, 2020

work page 2020
[2]

Semiconductor final testing scheduling using Q -learning based hyper -heuristic,

J. Lin, Y .-Y . Li, and H. -B. Song, “Semiconductor final testing scheduling using Q -learning based hyper -heuristic,” Expert Systems with Applications, vol. 187, pp. 115978, 2022

work page 2022
[3]

Modelling of connections and lateral behavior of high -rise modular steel buildings,

Y . Chua, J. R. Liew, and S. Pang, “Modelling of connections and lateral behavior of high -rise modular steel buildings,” Journal of Constructional Steel Research, vol. 166, pp. 105901, 2020

work page 2020
[4]

Structural materials: Metallurgy of bridges,

K. Ono, "Structural materials: Metallurgy of bridges," Metallurgical Design and Industry, pp. 193-269: Springer, 2018

work page 2018
[5]

Steel's role in future mobility,

A. Zoryk, “Steel's role in future mobility,” Steel Times International, vol. 43, no. 7, pp. 9-11, 2019

work page 2019
[6]

Hybrid scatter search algorithm for optimal and energy -efficient steelmaking -continuous casting,

Y . Tan, M. Zhou, Y . Zhang et al., “Hybrid scatter search algorithm for optimal and energy -efficient steelmaking -continuous casting,” IEEE Transactions on Automation Science and Engineering, vol. 17, no. 4, pp. 1814-1828, 2020

work page 2020
[7]

Research on Decision Support Method for Charge Batch Planning of Steelmaking -Continuous Casting under Lagrangian Framework,

L. Sun, Y . Y u, C. Li et al., “Research on Decision Support Method for Charge Batch Planning of Steelmaking -Continuous Casting under Lagrangian Framework,” IF AC-PapersOnLine, vol. 53, no. 2, pp. 11138-11143, 2020

work page 2020
[8]

An improved artificial bee colony algorithm for real-world hybrid flowshop rescheduling in steelmaking- refining-continuous casting process,

K. Peng, Q.-K. Pan, L. Gao et al., “An improved artificial bee colony algorithm for real-world hybrid flowshop rescheduling in steelmaking- refining-continuous casting process,” Computers & Industrial Engineering, vol. 122, pp. 235-250, 2018

work page 2018

Show all 37 references

[9]

A hybrid flowshop scheduling problem for a cold treating process in seamless steel tube production,

L. Li, J. Huo, and O. Tang, “A hybrid flowshop scheduling problem for a cold treating process in seamless steel tube production,” International journal of production research, vol. 49, no. 15, pp. 4679-4700, 2011

2011
[10]

An effective co -evolutionary artificial bee colony algorithm for steelmaking -continuous casting scheduling,

Q.-K. Pan, “An effective co -evolutionary artificial bee colony algorithm for steelmaking -continuous casting scheduling,” European Journal of Operational Research, vol. 250, no. 3, pp. 702-714, 2016

2016
[11]

Schrijver, Combinatorial optimization: polyhedra and efficiency: Springer, 2003

A. Schrijver, Combinatorial optimization: polyhedra and efficiency: Springer, 2003

2003
[12]

D.-S. Chen, R. G. Batson, and Y . Dang, Applied integer programming: modeling and solution: John Wiley & Sons, 2011

2011
[13]

A survey on new generation metaheuristic algorithms,

T. Dokeroglu, E. Sevinc, T. Kucukyilmaz et al., “A survey on new generation metaheuristic algorithms,” Computers & Industrial Engineering, vol. 137, pp. 106040, 2019

2019
[14]

Metaheuristics: A bibliography,

I. H. Osman, and G. Laporte, "Metaheuristics: A bibliography," Springer, 1996

1996
[15]

Metaheuristic algorithms: A comprehensive review,

M. Abdel -Basset, L. Abdel -Fatah, and A. K. San gaiah, “Metaheuristic algorithms: A comprehensive review,” Computational intelligence for multimedia big data on the cloud with engineering applications, pp. 185-231, 2018

2018
[16]

Multi-objective evolutionary algorithm based on multiple neighborhoods local search for multi -objective distributed hybrid flow shop scheduling problem,

W. Shao, Z. Shao, and D. Pi, “Multi-objective evolutionary algorithm based on multiple neighborhoods local search for multi -objective distributed hybrid flow shop scheduling problem,” Expert Systems with Applications, pp. 115453, 2021

2021
[17]

A copula -based hybrid estimation of distribution algorithm for m-machine reentrant permutation flow -shop scheduling problem,

B. Qian, Z.-c. Li, and R. Hu, “A copula -based hybrid estimation of distribution algorithm for m-machine reentrant permutation flow -shop scheduling problem,” Applied Soft Computing, vol. 61, pp. 921 -934, 2017

2017
[18]

Discrete evolutionary multi -objective optimization for energy -efficient blocking flow shop scheduling with setup time,

Han, J. Li, H. Sang et al. , “Discrete evolutionary multi -objective optimization for energy -efficient blocking flow shop scheduling with setup time,” Applied Soft Computing, vol. 93, pp. 106343, 2020

2020
[19]

Variable neighborhood search,

N. Mladenović, and P. Hansen, “Variable neighborhood search,” Computers & operations research, vol. 24, no. 11, pp. 1097-1100, 1997

1997
[20]

Iterated local search: Framework and applications,

H. R. Lourenço, O. C. Martin, and T. Stützle, "Iterated local search: Framework and applications," Handbook of metaheuristics, pp. 129-168: Springer, 2019

2019
[21]

Large neighborhood search,

D. Pisinger, and S. Ropke, "Large neighborhood search," Handbook of metaheuristics, pp. 99-127: Springer, 2019

2019
[22]

A genetic programming hyper-heuristic for the distributed assembly permutation flow -shop scheduling problem with sequence dependent setup times,

H.-B. Song, and J. Lin, “A genetic programming hyper-heuristic for the distributed assembly permutation flow -shop scheduling problem with sequence dependent setup times,” Swarm and Evolutionary Computation, vol. 60, pp. 100807, 2021

2021
[23]

A Review of Reinforcement Learning Based Intelligent Optimization for Manufacturing Scheduling,

L. Wang, Z. Pan, and J. Wang, “A Review of Reinforcement Learning Based Intelligent Optimization for Manufacturing Scheduling,” Complex System Modeling and Simulation, vol. 1, no. 4, pp. 257-270, 2021

2021
[24]

Automatic design of hyper-heuristic based on reinforcement learning,

S. S. Choong, L. -P. Wong, and C. P. Lim, “Automatic design of hyper-heuristic based on reinforcement learning,” Information Sciences, vol. 436, pp. 89-107, 2018

2018
[25]

Designing reusable metaheuristic methods: A semi-automated approach

S. Adriaensen, T. Brys, and A. Nowé, "Designing reusable metaheuristic methods: A semi-automated approach." pp. 2969-2976
[26]

Co -evolution Cross -entropy Optimization Algorithm for Cast Uncertain Steelmaking -continuous Casting Scheduling,

Y . Lv, B. Qian, R. Hu et al. , “Co -evolution Cross -entropy Optimization Algorithm for Cast Uncertain Steelmaking -continuous Casting Scheduling,” Journal of Mechanical Engineering, vol. 57, no. 19, pp. 192-207, 2021-10-05, 2021

2021
[27]

Fuzzy clustering with Gaussian-type member ship function

C. R. Ramesh, G. Jena, K. R. Rao et al. , "Fuzzy clustering with Gaussian-type member ship function." pp. 393-396
[28]

J.-J. Ji, Y . -N. Guo, X. -Z. Gao et al. , “Q -Learning-Based Hyperheuristic Evolutionary Algorithm for Dynamic Task Allocation of HierC_Q CCABC IABC EIGA IPSO HFOA HGA PIDEISA 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0 Average ARPD λ 200 300 400 HierC_Q CCABC IABC EIGA IPSO HFOA HGA PID...

2021
[29]

A self -learning genetic algorithm based on reinforcement learning for flexible job -shop scheduling problem,

R. Chen, B. Yang, S. Li et al. , “A self -learning genetic algorithm based on reinforcement learning for flexible job -shop scheduling problem,” Computers & Industrial Engineering, vol. 149, pp. 106778, 2020

2020
[30]

An estimation of distribution algorithm- based memetic algorithm for the distributed assembly permutation flow- shop scheduling problem,

S.-Y . Wang, and L. Wang, “An estimation of distribution algorithm- based memetic algorithm for the distributed assembly permutation flow- shop scheduling problem,” IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 46, no. 1, pp. 139-149, 2016

2016
[31]

A great deluge algorithm for a real-world examination timetabling problem,

M. Mohmad Kahar, and G. Kendall, “A great deluge algorithm for a real-world examination timetabling problem,” Journal of the Operational Research Society, vol. 66, no. 1, pp. 116-133, 2015

2015
[32]

An improved differential evolution algorithm for practical dynamic scheduling in steelmaking -continuous casting production,

L. Tang, Y . Zhao, and J. Liu, “An improved differential evolution algorithm for practical dynamic scheduling in steelmaking -continuous casting production,” IEEE Transactions on Evolutionary Computation, vol. 18, no. 2, pp. 209-225, 2014, 2014

2014
[33]

A hybrid fruit fly optimization algorithm for the realistic hybrid flowshop rescheduling problem in steelmaking systems,

Li, Q. Pan, and K. Mao , “A hybrid fruit fly optimization algorithm for the realistic hybrid flowshop rescheduling problem in steelmaking systems,” IEEE Transactions on Automation Science and Engineering, vol. 13, no. 2, pp. 932-949, 2016

2016
[34]

An improved simulated annealing for hybrid flowshops with sequence -dependent setup and transportation times to minimize total completion time and total tardiness,

B. Naderi, M. Zandieh, A. K. G. Ba lagh et al. , “An improved simulated annealing for hybrid flowshops with sequence -dependent setup and transportation times to minimize total completion time and total tardiness,” Expert systems with Applications, vol. 36, no. 6, pp. 9625-9633, 2009

2009
[35]

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

R. Ruiz, and C. Maroto, “A genetic algorithm for hybrid flowshops with sequence dependent setup times and machine eligibility,” European journal of operational research, vol. 169, no. 3, pp. 781-800, 2006

2006
[36]

An im proved particle swarm optimization algorithm to solve hybrid flowshop scheduling problems with the effect of human factors–A case study,

M. Marichelvam, M. Geetha, and Ö. Tosun, “An im proved particle swarm optimization algorithm to solve hybrid flowshop scheduling problems with the effect of human factors–A case study,” Computers & Operations Research, vol. 114, pp. 104812, 2020

2020
[37]

An improved iterated greedy algorithm for the energy-efficient blocking hybrid flow shop scheduling problem,

H.-X. Qin, Y .-Y . Han, B. Zhang et al., “An improved iterated greedy algorithm for the energy-efficient blocking hybrid flow shop scheduling problem,” Swarm and Evolutionary Computation, pp. 100992, 2021

2021

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Production scheduling in dynamic real-time optimization with closed-loop prediction,

J. E. Remigio, and C. L. Swartz, “Production scheduling in dynamic real-time optimization with closed-loop prediction,” Journal of Process Control, vol. 89, pp. 95-107, 2020

work page 2020

[2] [2]

Semiconductor final testing scheduling using Q -learning based hyper -heuristic,

J. Lin, Y .-Y . Li, and H. -B. Song, “Semiconductor final testing scheduling using Q -learning based hyper -heuristic,” Expert Systems with Applications, vol. 187, pp. 115978, 2022

work page 2022

[3] [3]

Modelling of connections and lateral behavior of high -rise modular steel buildings,

Y . Chua, J. R. Liew, and S. Pang, “Modelling of connections and lateral behavior of high -rise modular steel buildings,” Journal of Constructional Steel Research, vol. 166, pp. 105901, 2020

work page 2020

[4] [4]

Structural materials: Metallurgy of bridges,

K. Ono, "Structural materials: Metallurgy of bridges," Metallurgical Design and Industry, pp. 193-269: Springer, 2018

work page 2018

[5] [5]

Steel's role in future mobility,

A. Zoryk, “Steel's role in future mobility,” Steel Times International, vol. 43, no. 7, pp. 9-11, 2019

work page 2019

[6] [6]

Hybrid scatter search algorithm for optimal and energy -efficient steelmaking -continuous casting,

Y . Tan, M. Zhou, Y . Zhang et al., “Hybrid scatter search algorithm for optimal and energy -efficient steelmaking -continuous casting,” IEEE Transactions on Automation Science and Engineering, vol. 17, no. 4, pp. 1814-1828, 2020

work page 2020

[7] [7]

Research on Decision Support Method for Charge Batch Planning of Steelmaking -Continuous Casting under Lagrangian Framework,

L. Sun, Y . Y u, C. Li et al., “Research on Decision Support Method for Charge Batch Planning of Steelmaking -Continuous Casting under Lagrangian Framework,” IF AC-PapersOnLine, vol. 53, no. 2, pp. 11138-11143, 2020

work page 2020

[8] [8]

An improved artificial bee colony algorithm for real-world hybrid flowshop rescheduling in steelmaking- refining-continuous casting process,

K. Peng, Q.-K. Pan, L. Gao et al., “An improved artificial bee colony algorithm for real-world hybrid flowshop rescheduling in steelmaking- refining-continuous casting process,” Computers & Industrial Engineering, vol. 122, pp. 235-250, 2018

work page 2018