Swarm and Evolutionary Computation最新文献_第8页

Evolutionary Neural Architecture Search for Super-Resolution: Benchmarking SynFlow and model-based predictors 超分辨率的进化神经架构搜索：基准SynFlow和基于模型的预测器

IF 8.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Swarm and Evolutionary Computation

Pub Date : 2025-11-29 DOI: 10.1016/j.swevo.2025.102236

Sergio Sarmiento-Rosales, Víctor Adrían Sosa Hernández, Raúl Monroy

Neural Architecture Search (NAS) has emerged as a powerful tool for designing deep learning networks, particularly in image classification tasks. However, its application to Super-Resolution (SR) remains computationally demanding due to the high cost of architecture evaluations. To address this challenge, this study proposed a multi-objective NAS approach for SR, comparing performance predictors. Specifically, SynFlow serves as a zero-shot approach, providing a baseline for rapid, training-free performance estimation. At the same time, model-based predictors such as Support Vector Regression, Extra Trees, and XGBoost provide learned estimations from a subset of partially trained architectures. We employ the Non-Dominated Sorting Genetic Algorithm-III (NSGA-III) to search for architectures that maximize Peak Signal-to-Noise Ratio (PSNR) while minimizing computational complexity, measured in terms of parameter number and FLOPs. Our findings reveal that model-based methods yield more accurate performance estimations, resulting in better-balanced and more evenly distributed solutions across the Pareto front. This study contributes to the advancement of efficient NAS methodologies for SR applications by highlighting the trade-offs between computational efficiency and predictive similarity in evaluation methods.

神经结构搜索（NAS）已经成为设计深度学习网络的强大工具，特别是在图像分类任务中。然而，由于架构评估的高成本，它在超分辨率（SR）中的应用仍然需要大量的计算。为了应对这一挑战，本研究提出了一种多目标NAS方法来比较SR的性能预测指标。具体来说，SynFlow可以作为一种零射击方法，为快速、无需训练的性能评估提供基线。与此同时，基于模型的预测器，如支持向量回归、额外树和XGBoost，从部分训练的架构子集中提供学习估计。我们采用非支配排序遗传算法- iii （NSGA-III）来搜索最大化峰值信噪比（PSNR）同时最小化计算复杂性的架构，以参数数量和FLOPs来衡量。我们的研究结果表明，基于模型的方法产生更准确的性能估计，从而在整个帕累托前沿产生更好的平衡和更均匀分布的解决方案。本研究通过强调评估方法中计算效率和预测相似性之间的权衡，有助于提高高效的NAS方法在SR应用中的应用。

{"title":"Evolutionary Neural Architecture Search for Super-Resolution: Benchmarking SynFlow and model-based predictors","authors":"Sergio Sarmiento-Rosales, Víctor Adrían Sosa Hernández, Raúl Monroy","doi":"10.1016/j.swevo.2025.102236","DOIUrl":"10.1016/j.swevo.2025.102236","url":null,"abstract":"<div><div>Neural Architecture Search (NAS) has emerged as a powerful tool for designing deep learning networks, particularly in image classification tasks. However, its application to Super-Resolution (SR) remains computationally demanding due to the high cost of architecture evaluations. To address this challenge, this study proposed a multi-objective NAS approach for SR, comparing performance predictors. Specifically, SynFlow serves as a zero-shot approach, providing a baseline for rapid, training-free performance estimation. At the same time, model-based predictors such as Support Vector Regression, Extra Trees, and XGBoost provide learned estimations from a subset of partially trained architectures. We employ the Non-Dominated Sorting Genetic Algorithm-III (NSGA-III) to search for architectures that maximize Peak Signal-to-Noise Ratio (PSNR) while minimizing computational complexity, measured in terms of parameter number and FLOPs. Our findings reveal that model-based methods yield more accurate performance estimations, resulting in better-balanced and more evenly distributed solutions across the Pareto front. This study contributes to the advancement of efficient NAS methodologies for SR applications by highlighting the trade-offs between computational efficiency and predictive similarity in evaluation methods.</div></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"100 ","pages":"Article 102236"},"PeriodicalIF":8.5,"publicationDate":"2025-11-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145616108","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A novel encoding/decoding algorithm for constraint handling in a horizontal-axis wind turbine blade optimization process 一种新的水平轴风力机叶片优化过程约束处理编解码算法

IF 8.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Swarm and Evolutionary Computation

Pub Date : 2025-11-29 DOI: 10.1016/j.swevo.2025.102235

Krittatee Sangounsak, Sujin Bureerat, Akraphon Janon

The aim of this research is to improve the speed of the wind turbine blade optimization process while consuming less computation power. The traditional process using the constraint optimization technique and metaheuristics (MHs) cannot simultaneously handle many design variables, which is usually addressed by simplifying the problem. In this study, a novel encoding/decoding algorithm for constraint handling (ENDEACH) was constructed to handle all design variables prior to the application of MHs. The encoding process introduces the decision variable vector containing random numbers and boundaries for all design variables. Then the decoding process generates design variables from the vector. The technique produces design variables that are random while being within technical boundaries. Therefore, unfeasible search areas were eliminated. The proposed hybrid optimization process was applied to a 5-MW horizontal-axis wind turbine blade design. 53 design variables, including chord length, twist angle, chord distribution slope, twist distribution slope, and airfoil shape, were treated by ENDEACH. Then, several up-to-date MHs were used to optimize for the maximum power coefficient of wind turbines. The performance gains of the MHs were indicated by the highest convergence rate and minimum fitness value. Moreover, the optimal rotor had a power coefficient of up to 0.4987, 7.28 % higher than that of the National Renewable Energy Laboratory’s 5-MW reference wind turbine. The encoding/decoding algorithm together with MHs significantly outperformed the traditional method with respect to the optimization process and wind turbine blade performance, thereby enabling us to complete the optimization more quickly and obtain better results.

本研究的目的是提高风力机叶片优化过程的速度，同时减少计算量的消耗。传统的约束优化和元启发式方法不能同时处理多个设计变量，通常通过简化问题来解决这一问题。在本研究中，构建了一种新的约束处理编码/解码算法（ENDEACH）来处理mh应用前的所有设计变量。编码过程引入了包含随机数和所有设计变量边界的决策变量向量。然后解码过程从向量中生成设计变量。该技术在技术范围内产生随机的设计变量。因此，排除了不可行的搜索区域。将所提出的混合优化方法应用于5mw水平轴风力机叶片设计。采用ENDEACH对53个设计变量进行了处理，包括弦长、扭角、弦分布斜率、扭分布斜率和翼型形状。然后，利用几个最新的模型对风力发电机的最大功率系数进行了优化。最高的收敛速度和最小的适应度值表明了mhh的性能提升。最优转子功率系数高达0.4987，比国家可再生能源实验室5mw参考风力机功率系数提高7.28%。编/解码算法结合MHs在优化过程和风力机叶片性能上都明显优于传统方法，使我们能够更快地完成优化并获得更好的结果。

{"title":"A novel encoding/decoding algorithm for constraint handling in a horizontal-axis wind turbine blade optimization process","authors":"Krittatee Sangounsak, Sujin Bureerat, Akraphon Janon","doi":"10.1016/j.swevo.2025.102235","DOIUrl":"10.1016/j.swevo.2025.102235","url":null,"abstract":"<div><div>The aim of this research is to improve the speed of the wind turbine blade optimization process while consuming less computation power. The traditional process using the constraint optimization technique and metaheuristics (MHs) cannot simultaneously handle many design variables, which is usually addressed by simplifying the problem. In this study, a novel encoding/decoding algorithm for constraint handling (ENDEACH) was constructed to handle all design variables prior to the application of MHs. The encoding process introduces the decision variable vector containing random numbers and boundaries for all design variables. Then the decoding process generates design variables from the vector. The technique produces design variables that are random while being within technical boundaries. Therefore, unfeasible search areas were eliminated. The proposed hybrid optimization process was applied to a 5-MW horizontal-axis wind turbine blade design. 53 design variables, including chord length, twist angle, chord distribution slope, twist distribution slope, and airfoil shape, were treated by ENDEACH. Then, several up-to-date MHs were used to optimize for the maximum power coefficient of wind turbines. The performance gains of the MHs were indicated by the highest convergence rate and minimum fitness value. Moreover, the optimal rotor had a power coefficient of up to 0.4987, 7.28 % higher than that of the National Renewable Energy Laboratory’s 5-MW reference wind turbine. The encoding/decoding algorithm together with MHs significantly outperformed the traditional method with respect to the optimization process and wind turbine blade performance, thereby enabling us to complete the optimization more quickly and obtain better results.</div></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"100 ","pages":"Article 102235"},"PeriodicalIF":8.5,"publicationDate":"2025-11-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145684937","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A survey on Quality-Diversity optimization: Approaches, applications, and challenges 质量多样性优化研究进展：方法、应用和挑战

IF 8.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Swarm and Evolutionary Computation

Pub Date : 2025-11-28 DOI: 10.1016/j.swevo.2025.102240

Haoxiang Qin , Yi Xiang , Hainan Zhang , Yuyan Han , Yuting Wang , Xinrui Tao , Yiping Liu

Quality-Diversity (QD) optimization is a paradigm of evolutionary computation (EC) that extends the classic approaches, aiming to generate a collection of solutions that are both diverse and high-performing. Unlike traditional evolutionary algorithms (EAs), QD methods emphasize the illumination (or coverage) of a user-defined feature space, while simultaneously aiming for local optimization within each discovered region of the feature space. Over the past decade, QD has rapidly developed and proven effective in areas such as evolutionary robotics and video games. However, a systematic review of this growing field remains lacking. To date, the most recent review article on QD was published in 2021. Therefore, to offer a more comprehensive overview of the latest QD research, this paper provides a thorough survey of QD optimization, covering its foundational principles and representative algorithmic frameworks such as Novelty Search with Local Competition (NSLC), MAP-Elites, the unified modular QD framework, and RIBS. In addition, we divide the algorithm improvement part into three modules for discussion: containers, selection, and mutation. Then, the evaluation metrics widely used in QD optimization are listed for researchers. We further explore its diverse applications across domains such as evolutionary robotics, video games, scheduling, software testing, and engineering design. Finally, we discuss the current challenges in the field and outline promising directions for future research.

质量多样性（QD）优化是进化计算（EC）的一个范例，它扩展了经典方法，旨在生成既多样又高性能的解决方案集合。与传统的进化算法（EAs）不同，QD方法强调用户定义的特征空间的光照（或覆盖），同时在每个发现的特征空间区域内进行局部优化。在过去的十年中，量子点快速发展，并在进化机器人和电子游戏等领域被证明是有效的。然而，对这一不断发展的领域的系统回顾仍然缺乏。到目前为止，关于量子点的最新评论文章发表于2021年。因此，为了更全面地概述量子点优化的最新研究成果，本文对量子点优化的基本原理和具有代表性的算法框架（如新颖性搜索与局部竞争（NSLC）、MAP-Elites、统一模块化量子点优化框架和肋骨）进行了全面的综述。此外，我们将算法改进部分分为容器、选择和变异三个模块进行讨论。然后，列举了QD优化中常用的评价指标，供研究者参考。我们进一步探讨了它在进化机器人、视频游戏、调度、软件测试和工程设计等领域的各种应用。最后，我们讨论了该领域目前面临的挑战，并概述了未来研究的前景。

{"title":"A survey on Quality-Diversity optimization: Approaches, applications, and challenges","authors":"Haoxiang Qin , Yi Xiang , Hainan Zhang , Yuyan Han , Yuting Wang , Xinrui Tao , Yiping Liu","doi":"10.1016/j.swevo.2025.102240","DOIUrl":"10.1016/j.swevo.2025.102240","url":null,"abstract":"<div><div>Quality-Diversity (QD) optimization is a paradigm of evolutionary computation (EC) that extends the classic approaches, aiming to generate a collection of solutions that are both diverse and high-performing. Unlike traditional evolutionary algorithms (EAs), QD methods emphasize the illumination (or coverage) of a user-defined feature space, while simultaneously aiming for local optimization within each discovered region of the feature space. Over the past decade, QD has rapidly developed and proven effective in areas such as evolutionary robotics and video games. However, a systematic review of this growing field remains lacking. To date, the most recent review article on QD was published in 2021. Therefore, to offer a more comprehensive overview of the latest QD research, this paper provides a thorough survey of QD optimization, covering its foundational principles and representative algorithmic frameworks such as Novelty Search with Local Competition (NSLC), MAP-Elites, the unified modular QD framework, and RIBS. In addition, we divide the algorithm improvement part into three modules for discussion: containers, selection, and mutation. Then, the evaluation metrics widely used in QD optimization are listed for researchers. We further explore its diverse applications across domains such as evolutionary robotics, video games, scheduling, software testing, and engineering design. Finally, we discuss the current challenges in the field and outline promising directions for future research.</div></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"100 ","pages":"Article 102240"},"PeriodicalIF":8.5,"publicationDate":"2025-11-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145616106","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A Multi-population genetic programming algorithm to model estimation for LLC resonant converter 基于多种群遗传规划的LLC谐振变换器模型估计

IF 8.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Swarm and Evolutionary Computation

Pub Date : 2025-11-27 DOI: 10.1016/j.swevo.2025.102198

Yuefeng Liao , Zhiqiang Wang , Ying Bi , Jing Liang , Xiao Han

An accurate model of the LLC resonant converter, serving as a critical energy conversion stage in new energy storage converters, plays a pivotal role in its optimized design. However, traditional modeling approaches often fail to achieve ideal performance in practical applications, particularly due to parasitic parameter effects that cause significant discrepancies between theoretical and actual voltage gain. To better capture the nonlinear characteristics of the circuit, this paper adopts a data-driven approach and introduces the Multi-Population Genetic Programming (MPGP) algorithm, which enhances model interpretability and clarity. MPGP employs a multi-population strategy to optimize the evolutionary process, improving search capability and ensuring the generation of more precise models even with limited training data. Experimental validation on four dataset groups of LLC resonant converter demonstrates that MPGP significantly outperforms first harmonic approximation (FHA), genetic programming (GP), and five state-of-the-art regression methods in terms of estimation accuracy, model visualization, and interpretability. Moreover, the algorithm strengthens support for power electronics converter design, contributing to improved optimization of power electronic systems.

LLC谐振变换器作为新型储能变换器中能量转换的关键环节，其准确的模型对其优化设计起着至关重要的作用。然而，传统的建模方法在实际应用中往往不能达到理想的性能，特别是由于寄生参数效应导致理论和实际电压增益之间存在显着差异。为了更好地捕捉电路的非线性特性，本文采用数据驱动的方法，引入了多种群遗传规划（MPGP）算法，提高了模型的可解释性和清晰度。MPGP采用多种群策略优化进化过程，提高搜索能力，确保在有限的训练数据下生成更精确的模型。在四组LLC谐振转换器数据集上的实验验证表明，MPGP在估计精度、模型可视化和可解释性方面显著优于一谐波近似（FHA）、遗传规划（GP）和五种最先进的回归方法。此外，该算法还加强了对电力电子变换器设计的支持，有助于电力电子系统的优化。

{"title":"A Multi-population genetic programming algorithm to model estimation for LLC resonant converter","authors":"Yuefeng Liao , Zhiqiang Wang , Ying Bi , Jing Liang , Xiao Han","doi":"10.1016/j.swevo.2025.102198","DOIUrl":"10.1016/j.swevo.2025.102198","url":null,"abstract":"<div><div>An accurate model of the LLC resonant converter, serving as a critical energy conversion stage in new energy storage converters, plays a pivotal role in its optimized design. However, traditional modeling approaches often fail to achieve ideal performance in practical applications, particularly due to parasitic parameter effects that cause significant discrepancies between theoretical and actual voltage gain. To better capture the nonlinear characteristics of the circuit, this paper adopts a data-driven approach and introduces the Multi-Population Genetic Programming (MPGP) algorithm, which enhances model interpretability and clarity. MPGP employs a multi-population strategy to optimize the evolutionary process, improving search capability and ensuring the generation of more precise models even with limited training data. Experimental validation on four dataset groups of LLC resonant converter demonstrates that MPGP significantly outperforms first harmonic approximation (FHA), genetic programming (GP), and five state-of-the-art regression methods in terms of estimation accuracy, model visualization, and interpretability. Moreover, the algorithm strengthens support for power electronics converter design, contributing to improved optimization of power electronic systems.</div></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"100 ","pages":"Article 102198"},"PeriodicalIF":8.5,"publicationDate":"2025-11-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145616138","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Surrogate-assisted evolutionary algorithm with adaptive local region search for high-dimensional expensive multi-objective optimization problems 高维昂贵多目标优化问题的自适应局部搜索代理辅助进化算法

IF 8.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Swarm and Evolutionary Computation

Pub Date : 2025-11-24 DOI: 10.1016/j.swevo.2025.102232

Qing Wang , Huijun Li , Wei Zhang , Yong Zhang , Dunwei Gong , Fei Chu , Ali Wagdy Mohamed , Ilyas Muhammad

High-dimensional expensive multi-objective optimization problems (HEMOPs) are ubiquitous in scientific and engineering domains. They pose significant challenges for surrogate-assisted evolutionary algorithms (SAEAs), primarily due to the curse of dimensionality and prohibitive computational costs. A common approach involves partitioning the high-dimensional search space into local regions potentially containing Pareto optimal solutions, followed by in-depth exploration using SAEAs. However, existing partition-based evolutionary methods are limited by the lack of adaptability of local regions to environmental changes. To overcome this limitation, we propose AS-SMEA, a Surrogate-assisted Multi-objective Evolutionary Algorithm with Adaptive local region Search. The algorithm dynamically identifies and partitions promising regions in the high-dimensional space, conducting parallel surrogate-assisted evolutionary searches for efficient cooperative optimization. It is enhanced by two key strategies: a Covariance Matrix Adaptation-based method for initializing and updating local regions, and a Multi-Armed Bandit-guided adaptive selection mechanism for balancing exploration and exploitation. Moreover, theoretical analysis based on cumulative hypervolume regret establishes the global convergence of AS-SMEA. Comprehensive experiments on 69 benchmark problems and one real-world very large-scale integration design flow problem demonstrate that AS-SMEA consistently outperforms six state-of-the-art SAEAs.

高维昂贵多目标优化问题（HEMOPs）在科学和工程领域中普遍存在。它们对代理辅助进化算法（saea）提出了重大挑战，主要是由于维度的诅咒和令人望而却步的计算成本。一种常见的方法是将高维搜索空间划分为可能包含帕累托最优解的局部区域，然后使用saea进行深入探索。然而，现有的基于分区的进化方法由于缺乏局部区域对环境变化的适应性而受到限制。为了克服这一限制，我们提出了一种具有自适应局部区域搜索的代理辅助多目标进化算法as - sma。该算法在高维空间中动态识别和划分有希望的区域，进行并行代理辅助进化搜索，实现高效的协同优化。基于协方差矩阵的局部区域初始化和更新方法，以及多臂强盗引导的平衡探索和开发的自适应选择机制，增强了该算法的有效性。此外，基于累积超容量后悔的理论分析证实了AS-SMEA的全局收敛性。对69个基准问题和一个现实世界的大规模集成设计流问题的综合实验表明，AS-SMEA始终优于6个最先进的saea。

{"title":"Surrogate-assisted evolutionary algorithm with adaptive local region search for high-dimensional expensive multi-objective optimization problems","authors":"Qing Wang , Huijun Li , Wei Zhang , Yong Zhang , Dunwei Gong , Fei Chu , Ali Wagdy Mohamed , Ilyas Muhammad","doi":"10.1016/j.swevo.2025.102232","DOIUrl":"10.1016/j.swevo.2025.102232","url":null,"abstract":"<div><div>High-dimensional expensive multi-objective optimization problems (HEMOPs) are ubiquitous in scientific and engineering domains. They pose significant challenges for surrogate-assisted evolutionary algorithms (SAEAs), primarily due to the curse of dimensionality and prohibitive computational costs. A common approach involves partitioning the high-dimensional search space into local regions potentially containing Pareto optimal solutions, followed by in-depth exploration using SAEAs. However, existing partition-based evolutionary methods are limited by the lack of adaptability of local regions to environmental changes. To overcome this limitation, we propose AS-SMEA, a Surrogate-assisted Multi-objective Evolutionary Algorithm with Adaptive local region Search. The algorithm dynamically identifies and partitions promising regions in the high-dimensional space, conducting parallel surrogate-assisted evolutionary searches for efficient cooperative optimization. It is enhanced by two key strategies: a Covariance Matrix Adaptation-based method for initializing and updating local regions, and a Multi-Armed Bandit-guided adaptive selection mechanism for balancing exploration and exploitation. Moreover, theoretical analysis based on cumulative hypervolume regret establishes the global convergence of AS-SMEA. Comprehensive experiments on 69 benchmark problems and one real-world very large-scale integration design flow problem demonstrate that AS-SMEA consistently outperforms six state-of-the-art SAEAs.</div></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"100 ","pages":"Article 102232"},"PeriodicalIF":8.5,"publicationDate":"2025-11-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145616110","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Deep reinforcement learning framework for the underground mine short-term production scheduling problem 地下矿山短期生产调度问题的深度强化学习框架

IF 8.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Swarm and Evolutionary Computation

Pub Date : 2025-11-24 DOI: 10.1016/j.swevo.2025.102229

Ning Li , Yi Ding , Bibo Dai , Liguan Wang , Jian Chang , Haiwang Ye , Qizhou Wang

The escalating global demand for industrial minerals necessitates efficient short-term production scheduling for large-scale multi-panel underground mines. This study presents an end-to-end deep reinforcement learning (DRL) framework that simultaneously optimizes block selection and equipment allocation to minimize stope extraction completion time. The scheduling problem is formulated as a Markov decision process (MDP) with domain-specific state features, a composite block–equipment action space, constraint-embedded transitions, and a heterogeneous graph environment. A graph attention network (GAT) is employed to extract state features, and proximal policy optimization (PPO) is trained on stochastic scheduling instances. Experimental results demonstrate that the proposed method consistently outperforms experience-based scheduling rules and metaheuristics across various problem scales, achieving a 14.1 % reduction in makespan and a 65-fold speedup over genetic algorithms in large-scale instances. The framework exhibits strong generalization, robustness, and real-time responsiveness, providing a viable solution for complex underground mine production scheduling.

全球对工业矿产需求的不断增长，要求大型多盘位地下矿山的短期生产调度必须高效。本研究提出了一个端到端深度强化学习（DRL）框架，该框架可以同时优化区块选择和设备分配，以最大限度地减少采场开采完成时间。该调度问题被表述为具有特定领域状态特征的马尔可夫决策过程（MDP）、复合块设备动作空间、约束嵌入转换和异构图环境。采用图关注网络（GAT）提取状态特征，并在随机调度实例上训练近端策略优化（PPO）。实验结果表明，该方法在各种问题尺度上都优于基于经验的调度规则和元启发式算法，在大规模实例中，完成时间减少14.1%，速度比遗传算法提高65倍。该框架具有较强的泛化性、鲁棒性和实时性，为复杂的地下矿山生产调度提供了可行的解决方案。

{"title":"Deep reinforcement learning framework for the underground mine short-term production scheduling problem","authors":"Ning Li , Yi Ding , Bibo Dai , Liguan Wang , Jian Chang , Haiwang Ye , Qizhou Wang","doi":"10.1016/j.swevo.2025.102229","DOIUrl":"10.1016/j.swevo.2025.102229","url":null,"abstract":"<div><div>The escalating global demand for industrial minerals necessitates efficient short-term production scheduling for large-scale multi-panel underground mines. This study presents an end-to-end deep reinforcement learning (DRL) framework that simultaneously optimizes block selection and equipment allocation to minimize stope extraction completion time. The scheduling problem is formulated as a Markov decision process (MDP) with domain-specific state features, a composite block–equipment action space, constraint-embedded transitions, and a heterogeneous graph environment. A graph attention network (GAT) is employed to extract state features, and proximal policy optimization (PPO) is trained on stochastic scheduling instances. Experimental results demonstrate that the proposed method consistently outperforms experience-based scheduling rules and metaheuristics across various problem scales, achieving a 14.1 % reduction in makespan and a 65-fold speedup over genetic algorithms in large-scale instances. The framework exhibits strong generalization, robustness, and real-time responsiveness, providing a viable solution for complex underground mine production scheduling.</div></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"100 ","pages":"Article 102229"},"PeriodicalIF":8.5,"publicationDate":"2025-11-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145616109","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A multi-objective hierarchical aggregation optimization algorithm for dynamic network structures in federated learning 联邦学习中动态网络结构的多目标分层聚合优化算法

IF 8.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Swarm and Evolutionary Computation

Pub Date : 2025-11-24 DOI: 10.1016/j.swevo.2025.102231

Wubin Ma , Kangqi Hu , Rui Wang , Qinming Sun , Yahui Wu , Chaofan Dai , Haohou Zhou , Martha Arbayani Zaidan

Federated learning facilitates big data analysis and applications while safeguarding user privacy, thus emerging as a pivotal paradigm in distributed machine learning. Building on this foundation, multi-objective federated learning (MOFL) — which focuses on the collaborative optimization of multiple objectives such as communication cost and computational efficiency — has become one of the current research hotspots. However, MOFL still suffers from performance degradation in scenarios with heterogeneous data and dynamic network topologies, limiting its practical applicability. To address these challenges, this paper proposes a multi-objective hierarchical aggregation optimization method tailored for dynamic network structures. Specifically, a hierarchical aggregation mechanism is adopted to tackle the dynamic variations in client-side neural network models, which optimizes the training process of MOFL and significantly enhances computational efficiency under dynamic network and heterogeneous data environments. Experimental results verify that the proposed method achieves remarkable performance improvements across different data distributions: it attains an average performance enhancement of 50.73% compared with the NSGA-III algorithm. Furthermore, comprehensive comparative experiments with other state-of-the-art multi-objective optimization algorithms demonstrate its overall superior performance, confirming the scalability of the proposed method in practical scenarios.

联邦学习促进了大数据分析和应用，同时保护了用户隐私，因此成为分布式机器学习的关键范式。在此基础上，多目标联邦学习（MOFL）成为当前的研究热点之一，其重点是通信成本和计算效率等多目标的协同优化。然而，在异构数据和动态网络拓扑的场景下，MOFL仍然存在性能下降的问题，限制了其实际应用。为了解决这些问题，本文提出了一种针对动态网络结构的多目标分层聚合优化方法。具体而言，采用层次聚合机制解决客户端神经网络模型的动态变化，优化了MOFL的训练过程，显著提高了动态网络和异构数据环境下的计算效率。实验结果表明，该方法在不同的数据分布下都取得了显著的性能提升，与NSGA-III算法相比，平均性能提升了50.73%。通过与其他多目标优化算法的综合对比实验，验证了该方法在实际场景中的可扩展性。

{"title":"A multi-objective hierarchical aggregation optimization algorithm for dynamic network structures in federated learning","authors":"Wubin Ma , Kangqi Hu , Rui Wang , Qinming Sun , Yahui Wu , Chaofan Dai , Haohou Zhou , Martha Arbayani Zaidan","doi":"10.1016/j.swevo.2025.102231","DOIUrl":"10.1016/j.swevo.2025.102231","url":null,"abstract":"<div><div>Federated learning facilitates big data analysis and applications while safeguarding user privacy, thus emerging as a pivotal paradigm in distributed machine learning. Building on this foundation, multi-objective federated learning (MOFL) — which focuses on the collaborative optimization of multiple objectives such as communication cost and computational efficiency — has become one of the current research hotspots. However, MOFL still suffers from performance degradation in scenarios with heterogeneous data and dynamic network topologies, limiting its practical applicability. To address these challenges, this paper proposes a multi-objective hierarchical aggregation optimization method tailored for dynamic network structures. Specifically, a hierarchical aggregation mechanism is adopted to tackle the dynamic variations in client-side neural network models, which optimizes the training process of MOFL and significantly enhances computational efficiency under dynamic network and heterogeneous data environments. Experimental results verify that the proposed method achieves remarkable performance improvements across different data distributions: it attains an average performance enhancement of 50.73% compared with the NSGA-III algorithm. Furthermore, comprehensive comparative experiments with other state-of-the-art multi-objective optimization algorithms demonstrate its overall superior performance, confirming the scalability of the proposed method in practical scenarios.</div></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"100 ","pages":"Article 102231"},"PeriodicalIF":8.5,"publicationDate":"2025-11-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145616105","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A constrained multi-objective dynamic corridor allocation problem model for low-carbon manufacturing 低碳制造的约束多目标动态走廊配置问题模型

IF 8.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Swarm and Evolutionary Computation

Pub Date : 2025-11-21 DOI: 10.1016/j.swevo.2025.102221

Zihan Guo , Zeqiang Zhang , Dan Ji , Zongxing He , Lei Guo , Tong Tian , Silu Liu

A well-designed facility layout can effectively reduce carbon emissions by optimising logistics and energy use. Dynamic environments enable more accurate emission estimation, they have rarely been considered in low-carbon manufacturing layout optimisation research. To address this gap, this study proposes an extended constrained multi-objective dynamic corridor allocation problem that integrates three optimisation objectives: logistics costs, carbon emissions, and layout space. A mixed-integer programming model incorporating carbon emissions was developed, considering the energy consumption of material handling equipment, and an estimation method for carbon emissions in a dynamic production environment was proposed. To more accurately represent energy use and spatial interactions, mathematical constraints were imposed to the facility positions, extending the model to a mixed-integer quadratic-constrained programming framework. Given the high computational complexity of the proposed problem, this study designed an improved non-dominated sorting genetic algorithm II (NSGA-II) that enhances the initial solution generation and refines the path-relinking strategy based on problem encoding and multi-objective characteristics, substantially enhance the solution convergence and diversity. Experimental results demonstrate that the proposed algorithm accurately solves four asymmetric material flow multi-objective dynamic instances and outperforms both the traditional NSGA-II and existing literature in benchmark dynamic instances by identifying superior Pareto-optimal solutions. In real-world cases it has been possible to optimise material handling costs by 23.43%, carbon costs by 7.9% and effectively improves shop floor space utilisation. This study provides a theoretical foundation and practical insights for advancing green manufacturing and low-carbon economies.

精心设计的设施布局可以通过优化物流和能源使用来有效减少碳排放。动态环境使排放估算更加准确，但在低碳制造布局优化研究中很少考虑。为了解决这一差距，本研究提出了一个扩展约束多目标动态走廊分配问题，该问题集成了三个优化目标：物流成本、碳排放和布局空间。考虑物料搬运设备的能耗，建立了考虑碳排放的混合整数规划模型，提出了动态生产环境下的碳排放估算方法。为了更准确地表示能源使用和空间相互作用，对设施位置施加了数学约束，将模型扩展为混合整数二次约束规划框架。针对所提问题的高计算复杂度，本研究设计了一种改进的非支配排序遗传算法II (NSGA-II)，增强了初始解生成，并基于问题编码和多目标特征对路径链接策略进行了细化，大大提高了解的收敛性和多样性。实验结果表明，该算法准确地求解了4个不对称物料流多目标动态实例，并通过识别优pareto最优解，在基准动态实例中优于传统NSGA-II和现有文献。在实际案例中，它可以优化23.43%的物料处理成本，7.9%的碳成本，并有效提高车间空间利用率。本研究为推动绿色制造和低碳经济发展提供了理论基础和实践启示。

{"title":"A constrained multi-objective dynamic corridor allocation problem model for low-carbon manufacturing","authors":"Zihan Guo , Zeqiang Zhang , Dan Ji , Zongxing He , Lei Guo , Tong Tian , Silu Liu","doi":"10.1016/j.swevo.2025.102221","DOIUrl":"10.1016/j.swevo.2025.102221","url":null,"abstract":"<div><div>A well-designed facility layout can effectively reduce carbon emissions by optimising logistics and energy use. Dynamic environments enable more accurate emission estimation, they have rarely been considered in low-carbon manufacturing layout optimisation research. To address this gap, this study proposes an extended constrained multi-objective dynamic corridor allocation problem that integrates three optimisation objectives: logistics costs, carbon emissions, and layout space. A mixed-integer programming model incorporating carbon emissions was developed, considering the energy consumption of material handling equipment, and an estimation method for carbon emissions in a dynamic production environment was proposed. To more accurately represent energy use and spatial interactions, mathematical constraints were imposed to the facility positions, extending the model to a mixed-integer quadratic-constrained programming framework. Given the high computational complexity of the proposed problem, this study designed an improved non-dominated sorting genetic algorithm II (NSGA-II) that enhances the initial solution generation and refines the path-relinking strategy based on problem encoding and multi-objective characteristics, substantially enhance the solution convergence and diversity. Experimental results demonstrate that the proposed algorithm accurately solves four asymmetric material flow multi-objective dynamic instances and outperforms both the traditional NSGA-II and existing literature in benchmark dynamic instances by identifying superior Pareto-optimal solutions. In real-world cases it has been possible to optimise material handling costs by 23.43%, carbon costs by 7.9% and effectively improves shop floor space utilisation. This study provides a theoretical foundation and practical insights for advancing green manufacturing and low-carbon economies.</div></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"100 ","pages":"Article 102221"},"PeriodicalIF":8.5,"publicationDate":"2025-11-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145571083","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Graph-based evolutionary search for maximum independent union of cliques problem 基于图的最大独立团并问题的进化搜索

IF 8.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Swarm and Evolutionary Computation

Pub Date : 2025-11-20 DOI: 10.1016/j.swevo.2025.102223

Zhenghao Xu , Ruixue Zhang , Xueshi Dong , Qihao Song

To address the maximum independent union of cliques (IUC) problem, this work introduces a Graph-based Evolutionary Search (GBES) algorithm. GBES utilizes the vertex connected status of the benchmark graph to prioritize vertices with the potential to form their own cliques. This effectively guides the convergence direction of the algorithm, thereby alleviating the local optima problem commonly found in existing methods. Also, this property allows GBES to be applied to other variants of the cliques problems or discrete problems. Specifically, GBES incorporates a graph-information-based crossover operator to generate offspring solutions and a constrained generalized exchange neighborhoods to enhance the search capability. Another strategy is a specific initialization method to ensure population diversification. Experimental results on 83 2nd DIMACS benchmark graphs indicate that the GBES competes favorably with the state-of-the-art algorithms. In particular, new lower bounds are found in 12 benchmark graphs and the best-known solutions are matched in the rest. The advantages of the key components of the algorithm are analyzed.

为了解决派系最大独立联合问题，本文引入了一种基于图的进化搜索算法（GBES）。GBES利用基准图的顶点连接状态来确定有可能形成自己的团的顶点的优先级。这有效地引导了算法的收敛方向，从而缓解了现有方法中常见的局部最优问题。此外，该特性还允许将GBES应用于团问题或离散问题的其他变体。具体来说，GBES结合了基于图信息的交叉算子生成子代解和约束广义交换邻域来增强搜索能力。另一种策略是特定的初始化方法，以确保种群多样化。在83个第二DIMACS基准图上的实验结果表明，GBES与最先进的算法具有良好的竞争优势。特别是，在12个基准图中发现了新的下界，并且在其余的图中匹配了最知名的解。分析了该算法各关键组成部分的优点。

{"title":"Graph-based evolutionary search for maximum independent union of cliques problem","authors":"Zhenghao Xu , Ruixue Zhang , Xueshi Dong , Qihao Song","doi":"10.1016/j.swevo.2025.102223","DOIUrl":"10.1016/j.swevo.2025.102223","url":null,"abstract":"<div><div>To address the maximum independent union of cliques (IUC) problem, this work introduces a Graph-based Evolutionary Search (GBES) algorithm. GBES utilizes the vertex connected status of the benchmark graph to prioritize vertices with the potential to form their own cliques. This effectively guides the convergence direction of the algorithm, thereby alleviating the local optima problem commonly found in existing methods. Also, this property allows GBES to be applied to other variants of the cliques problems or discrete problems. Specifically, GBES incorporates a graph-information-based crossover operator to generate offspring solutions and a constrained generalized exchange neighborhoods to enhance the search capability. Another strategy is a specific initialization method to ensure population diversification. Experimental results on 83 2nd DIMACS benchmark graphs indicate that the GBES competes favorably with the state-of-the-art algorithms. In particular, new lower bounds are found in 12 benchmark graphs and the best-known solutions are matched in the rest. The advantages of the key components of the algorithm are analyzed.</div></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"100 ","pages":"Article 102223"},"PeriodicalIF":8.5,"publicationDate":"2025-11-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145571082","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Particle swarm optimization based on K-means clustering and adaptive dual-groups strategy 基于k均值聚类和自适应双群策略的粒子群优化

IF 8.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Swarm and Evolutionary Computation

Pub Date : 2025-11-20 DOI: 10.1016/j.swevo.2025.102226

Yuyu Fan , Dongping Tian , Qinghao Xu , Jie Sun , Qiu Xu , Zhongzhi Shi

Particle swarm optimization (PSO) is a swarm intelligence algorithm that simulates the cooperative foraging behavior of bird flocks and searches for the optimal solution by iterating and updating the position and speed of particles. Its advantages are that the principle is simple and easy to implement, the convergence speed is fast, and it is suitable for high-dimensional problems. Nevertheless, it has the drawbacks of being prone to fall into local optimum, having low search efficiency in the later stage and relying on experience for parameter setting. Hence, this paper puts forward a particle swarm optimization algorithm based on k-means clustering and adaptive dual-groups strategy (PSO-KCAD) to solve the related problems mentioned above. First, a twin swarm collaborative search strategy is employed to co-evolve collaboratively and balance the exploration in the early stage of the search and the exploitation in the later stage. Second, comprehensive learning and subgroup elite-ordinary particle stratification strategy are used to promote communication among particles and thereby accelerate the convergence process. Subsequently, the adaptive probability-driven elite replacement and competitive disturbance mechanism are utilized to maintain population diversity and improve the accuracy of solutions. Finally, the performance of PSO-KCAD is compared with that of several other PSO variants on CEC2017. The experimental results show that PSO-KCAD is markedly superior to other algorithms. To further verify the effectiveness and robustness of our proposal, we apply it to two real-world problems and the results show that it has also achieved the most promising optimization results.

粒子群优化算法（Particle swarm optimization， PSO）是一种模拟鸟群协同觅食行为，通过迭代和更新粒子的位置和速度来寻找最优解的群体智能算法。其优点是原理简单，易于实现，收敛速度快，适用于高维问题。但存在容易陷入局部最优、后期搜索效率低、参数设置依赖经验等缺点。为此，本文提出了一种基于k均值聚类和自适应双群策略的粒子群优化算法（PSO-KCAD）来解决上述相关问题。首先，采用双群协同搜索策略进行协同进化，平衡搜索前期的探索和后期的开发；其次，采用综合学习和子群精英-普通粒子分层策略，促进粒子间的交流，从而加快收敛过程。然后，利用自适应概率驱动的精英替换和竞争扰动机制来保持种群多样性，提高解的准确性。最后，将PSO- kcad与其他几种PSO变体在CEC2017上的性能进行了比较。实验结果表明，PSO-KCAD算法明显优于其他算法。为了进一步验证我们的建议的有效性和鲁棒性，我们将其应用于两个现实问题，结果表明它也取得了最有希望的优化结果。

{"title":"Particle swarm optimization based on K-means clustering and adaptive dual-groups strategy","authors":"Yuyu Fan , Dongping Tian , Qinghao Xu , Jie Sun , Qiu Xu , Zhongzhi Shi","doi":"10.1016/j.swevo.2025.102226","DOIUrl":"10.1016/j.swevo.2025.102226","url":null,"abstract":"<div><div>Particle swarm optimization (PSO) is a swarm intelligence algorithm that simulates the cooperative foraging behavior of bird flocks and searches for the optimal solution by iterating and updating the position and speed of particles. Its advantages are that the principle is simple and easy to implement, the convergence speed is fast, and it is suitable for high-dimensional problems. Nevertheless, it has the drawbacks of being prone to fall into local optimum, having low search efficiency in the later stage and relying on experience for parameter setting. Hence, this paper puts forward a particle swarm optimization algorithm based on k-means clustering and adaptive dual-groups strategy (PSO-KCAD) to solve the related problems mentioned above. First, a twin swarm collaborative search strategy is employed to co-evolve collaboratively and balance the exploration in the early stage of the search and the exploitation in the later stage. Second, comprehensive learning and subgroup elite-ordinary particle stratification strategy are used to promote communication among particles and thereby accelerate the convergence process. Subsequently, the adaptive probability-driven elite replacement and competitive disturbance mechanism are utilized to maintain population diversity and improve the accuracy of solutions. Finally, the performance of PSO-KCAD is compared with that of several other PSO variants on CEC2017. The experimental results show that PSO-KCAD is markedly superior to other algorithms. To further verify the effectiveness and robustness of our proposal, we apply it to two real-world problems and the results show that it has also achieved the most promising optimization results.</div></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"100 ","pages":"Article 102226"},"PeriodicalIF":8.5,"publicationDate":"2025-11-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145571081","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0