Scholarlink

<p>Order picking is the most labor-intensive and cost-dominant core operational process in warehouse supply chains, and its efficiency level directly affects the overall responsiveness of logistics systems. To address the limitations of existing genetic algorithms in solving large-scale order picking optimization problems; including crude fitness evaluation, rigid parameter tuning, and insufficient initial population quality; this paper proposes an improved genetic algorithm driven by a digital twin–aligned simulation environment (DT-IGA). At the algorithmic level, this paper designs a hierarchical clustering population initialization strategy based on the spatial similarity of order storage locations, as well as an adaptive genetic operator mechanism that dynamically adjusts crossover and mutation rates according to population fitness distribution, to accelerate convergence while maintaining search diversity. In three sets of simulation experiments with different scales containing 50–300 orders, DT-IGA achieved optimization improvements of 14.7–17.2% in total picking time compared to the standard genetic algorithm, with convergence speed improved by approximately one-third, the average per-picker waiting time reduced by approximately 30–42% relative to the best benchmark, and significantly enhanced operational stability. Module ablation experiments verified the complementary and synergistic gain structure among the three modules: digital twin simulation evaluation, adaptive genetic operators, and clustering initialization. Experimental results demonstrate that embedding a digital twin–aligned simulation environment into the fitness evaluation loop of a genetic algorithm is an effective approach to improving the solution quality of order picking optimization. Statistical significance of the improvements is verified by Wilcoxon signed-rank tests on paired runs generated with a shared-seed protocol (all <i>p</i> &lt; 0.001, <i>W</i> ≤ 12 across the nine pairwise comparisons). Sensitivity analyses conducted on an independent held-out validation scenario further confirm robustness to ± 20% variations in walking speed, picking time, and key algorithm hyperparameters.</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

Intelligent warehouse supply chain order picking optimization based on digital twin and genetic algorithm

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月16日

查看全文
  • JingYun Zhang

摘要

Order picking is the most labor-intensive and cost-dominant core operational process in warehouse supply chains, and its efficiency level directly affects the overall responsiveness of logistics systems. To address the limitations of existing genetic algorithms in solving large-scale order picking optimization problems; including crude fitness evaluation, rigid parameter tuning, and insufficient initial population quality; this paper proposes an improved genetic algorithm driven by a digital twin–aligned simulation environment (DT-IGA). At the algorithmic level, this paper designs a hierarchical clustering population initialization strategy based on the spatial similarity of order storage locations, as well as an adaptive genetic operator mechanism that dynamically adjusts crossover and mutation rates according to population fitness distribution, to accelerate convergence while maintaining search diversity. In three sets of simulation experiments with different scales containing 50–300 orders, DT-IGA achieved optimization improvements of 14.7–17.2% in total picking time compared to the standard genetic algorithm, with convergence speed improved by approximately one-third, the average per-picker waiting time reduced by approximately 30–42% relative to the best benchmark, and significantly enhanced operational stability. Module ablation experiments verified the complementary and synergistic gain structure among the three modules: digital twin simulation evaluation, adaptive genetic operators, and clustering initialization. Experimental results demonstrate that embedding a digital twin–aligned simulation environment into the fitness evaluation loop of a genetic algorithm is an effective approach to improving the solution quality of order picking optimization. Statistical significance of the improvements is verified by Wilcoxon signed-rank tests on paired runs generated with a shared-seed protocol (all p < 0.001, W ≤ 12 across the nine pairwise comparisons). Sensitivity analyses conducted on an independent held-out validation scenario further confirm robustness to ± 20% variations in walking speed, picking time, and key algorithm hyperparameters.