Scholarlink

<p>Constructive heuristics constitute fundamental polynomial-time algorithms that systematically generate feasible solutions to NP-hard combinatorial optimization problems through incremental component assembly. This comprehensive systematic review analyzes 320 peer-reviewed articles published between January 2014 and June 2025, employing PRISMA methodology to provide the first cross-domain analysis of constructive heuristics across transportation, scheduling, and timetabling domains. Our analysis reveals pronounced research asymmetry: scheduling dominates with 282 papers (88.13%) while transportation and timetabling contribute equally with 18 papers each (5.63%), highlighting untapped cross-domain collaboration potential. Domain-specific algorithmic preferences reflect underlying problem characteristics: scheduling employs the Nawaz-Enscore-Ham heuristic (48 articles, 17.02%), transportation utilizes nearest-neighbor approaches (7 articles, 38.89%), and timetabling emphasizes graph-coloring-based construction (7 articles, 38.89%). Critically, constructive heuristics serve distinct functional roles across these domains—primary optimizer in scheduling, initialization seed in transportation, and feasibility generator in timetabling—justifying a unified cross-domain perspective. Integration patterns reveal sophisticated hybrid architectures with genetic algorithms achieving remarkable cross-domain consistency (22.3% scheduling, 50.0% transportation, 44.4% timetabling), while machine learning integration varies significantly (72.2% transportation, 50.0% timetabling, 32.3% scheduling). Critical evaluation deficiencies emerge: statistical testing adoption ranges from 24.1% (scheduling) to 66.7% (timetabling), with 67.5% of studies lacking validation entirely. Four fundamental limitation categories constrain advancement: computational barriers (scalability affecting 19.15% of scheduling research), methodological constraints (parameter sensitivity in 6.74% of studies), infrastructure deficiencies (67.5% lacking statistical validation), and contemporary adaptation challenges. Five strategic opportunities emerge: underexplored machine learning integration (19 neural network, 21 reinforcement learning studies), real-time adaptability for Industry 4.0, sustainability integration, cross-domain knowledge transfer, and explainability enhancement for critical applications. This review provides three contributions: cross-domain methodological synthesis with an explicit transferability mapping identifying which constructive principles transfer across domains and where transfer fails, a unified evaluation framework enabling evidence-based selection including a practitioner’s guide for method selection, and a strategic roadmap guiding systematic field advancement toward unified, scientifically rigorous optimization methodologies.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Constructive Heuristics for Combinatorial Optimization: A Comprehensive Cross-Domain Analysis of Algorithmic Methods, Integration Patterns, and Performance Characteristics

  • 期刊论文

  • 发表日期：

    2026年4月23日

查看全文
  • Masri Ayob,
  • Norizal Abdullah,
  • Saadeddin Hroub,
  • Oussama S. Enshassi,
  • Meng Chun Lam,
  • Nasser R. Sabar

摘要

Constructive heuristics constitute fundamental polynomial-time algorithms that systematically generate feasible solutions to NP-hard combinatorial optimization problems through incremental component assembly. This comprehensive systematic review analyzes 320 peer-reviewed articles published between January 2014 and June 2025, employing PRISMA methodology to provide the first cross-domain analysis of constructive heuristics across transportation, scheduling, and timetabling domains. Our analysis reveals pronounced research asymmetry: scheduling dominates with 282 papers (88.13%) while transportation and timetabling contribute equally with 18 papers each (5.63%), highlighting untapped cross-domain collaboration potential. Domain-specific algorithmic preferences reflect underlying problem characteristics: scheduling employs the Nawaz-Enscore-Ham heuristic (48 articles, 17.02%), transportation utilizes nearest-neighbor approaches (7 articles, 38.89%), and timetabling emphasizes graph-coloring-based construction (7 articles, 38.89%). Critically, constructive heuristics serve distinct functional roles across these domains—primary optimizer in scheduling, initialization seed in transportation, and feasibility generator in timetabling—justifying a unified cross-domain perspective. Integration patterns reveal sophisticated hybrid architectures with genetic algorithms achieving remarkable cross-domain consistency (22.3% scheduling, 50.0% transportation, 44.4% timetabling), while machine learning integration varies significantly (72.2% transportation, 50.0% timetabling, 32.3% scheduling). Critical evaluation deficiencies emerge: statistical testing adoption ranges from 24.1% (scheduling) to 66.7% (timetabling), with 67.5% of studies lacking validation entirely. Four fundamental limitation categories constrain advancement: computational barriers (scalability affecting 19.15% of scheduling research), methodological constraints (parameter sensitivity in 6.74% of studies), infrastructure deficiencies (67.5% lacking statistical validation), and contemporary adaptation challenges. Five strategic opportunities emerge: underexplored machine learning integration (19 neural network, 21 reinforcement learning studies), real-time adaptability for Industry 4.0, sustainability integration, cross-domain knowledge transfer, and explainability enhancement for critical applications. This review provides three contributions: cross-domain methodological synthesis with an explicit transferability mapping identifying which constructive principles transfer across domains and where transfer fails, a unified evaluation framework enabling evidence-based selection including a practitioner’s guide for method selection, and a strategic roadmap guiding systematic field advancement toward unified, scientifically rigorous optimization methodologies.