Sustainable Computing-Informatics & Systems最新文献_第3页

Sustainable machine learning-based routing in 5G-VANETs for reducing power consumption in real-time communications 5g - vanet中基于可持续机器学习的路由，用于降低实时通信中的功耗

IF 5.7 3区计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Sustainable Computing-Informatics & Systems

Pub Date : 2025-12-01 DOI: 10.1016/j.suscom.2025.101262

M. John Peter , R. Manoharan

Intelligent Transportation Systems (ITSs) are a critical application of Fifth-Generation (5 G) mobile communication technology, with Vehicular Ad Hoc Networks (VANETs) serving as a fundamental component. Although 5 G infrastructure significantly enhances connectivity, challenges persist in scenarios with limited coverage or high vehicle mobility, where Device-To-Device (D2D) communication becomes essential. VANETs further encounter unstable connectivity, rapidly changing topologies, and uneven vehicle distribution, which lead to frequent route rediscovery, excessive signaling overhead, and increased power consumption. To address these limitations, a sustainable machine learning (ML)-based routing protocol is developed that integrates 5 G and D2D communication for improved reliability and energy efficiency. This research utilizes a 5G-enabled VANET simulation environment to collect mobility, communication, and energy-related data for routing optimization. The dataset undergoes cleaning, standardization, and outlier detection to ensure reliability, while Wavelet Transform and PCA are applied for dimensionality reduction and pattern extraction. The Golden Jackal Optimization (GJO) algorithm is used for feature selection and parameter tuning, optimizing routing decisions and evaluating network connectivity through a nonhomogeneous Poisson process. Routing optimization is achieved using a novel ML model, the Extreme Kernelized Gradient Supported Machine (EKGSM), which exploits kernelized gradient learning to capture nonlinear mobility, connectivity, and energy-related patterns. The proposed model achieves an increase in packet delivery ratio (PDR), a reduction in average end-to-end (E2E) delay, a decrease in energy consumption, and a reduction in routing overhead. These outcomes establish EKGSM as an effective, scalable, and sustainable routing solution for next-generation 5G-VANET environments.

智能交通系统（its）是第五代（5 G）移动通信技术的关键应用，车辆自组织网络（VANETs）是一个基本组成部分。尽管5g 基础设施显著增强了连接性，但在覆盖范围有限或车辆移动性高的情况下，设备对设备（D2D）通信变得至关重要，挑战依然存在。此外，vanet还会遇到不稳定的连接、快速变化的拓扑结构和不均匀的车辆分布，从而导致频繁的路线重新发现、过多的信号开销和增加的功耗。为了解决这些限制，开发了一种基于可持续机器学习（ML）的路由协议，该协议集成了5 G和D2D通信，以提高可靠性和能源效率。该研究利用支持5g的VANET模拟环境来收集移动、通信和能源相关数据，以进行路由优化。对数据集进行清洗、标准化和离群点检测以保证数据的可靠性，同时利用小波变换和主成分分析进行降维和模式提取。Golden Jackal Optimization （GJO）算法用于特征选择和参数调整，优化路由决策，并通过非齐次泊松过程评估网络连通性。路由优化是使用一种新的机器学习模型，即极限核梯度支持机（EKGSM）来实现的，该模型利用核梯度学习来捕获非线性移动性、连通性和能量相关模式。该模型提高了分组传送率（PDR），降低了端到端平均延迟，降低了能耗，减少了路由开销。这些成果使EKGSM成为下一代5G-VANET环境中有效、可扩展和可持续的路由解决方案。

{"title":"Sustainable machine learning-based routing in 5G-VANETs for reducing power consumption in real-time communications","authors":"M. John Peter , R. Manoharan","doi":"10.1016/j.suscom.2025.101262","DOIUrl":"10.1016/j.suscom.2025.101262","url":null,"abstract":"<div><div>Intelligent Transportation Systems (ITSs) are a critical application of Fifth-Generation (5 G) mobile communication technology, with Vehicular Ad Hoc Networks (VANETs) serving as a fundamental component. Although 5 G infrastructure significantly enhances connectivity, challenges persist in scenarios with limited coverage or high vehicle mobility, where Device-To-Device (D2D) communication becomes essential. VANETs further encounter unstable connectivity, rapidly changing topologies, and uneven vehicle distribution, which lead to frequent route rediscovery, excessive signaling overhead, and increased power consumption. To address these limitations, a sustainable machine learning (ML)-based routing protocol is developed that integrates 5 G and D2D communication for improved reliability and energy efficiency. This research utilizes a 5G-enabled VANET simulation environment to collect mobility, communication, and energy-related data for routing optimization. The dataset undergoes cleaning, standardization, and outlier detection to ensure reliability, while Wavelet Transform and PCA are applied for dimensionality reduction and pattern extraction. The Golden Jackal Optimization (GJO) algorithm is used for feature selection and parameter tuning, optimizing routing decisions and evaluating network connectivity through a nonhomogeneous Poisson process. Routing optimization is achieved using a novel ML model, the Extreme Kernelized Gradient Supported Machine (EKGSM), which exploits kernelized gradient learning to capture nonlinear mobility, connectivity, and energy-related patterns. The proposed model achieves an increase in packet delivery ratio (PDR), a reduction in average end-to-end (E2E) delay, a decrease in energy consumption, and a reduction in routing overhead. These outcomes establish EKGSM as an effective, scalable, and sustainable routing solution for next-generation 5G-VANET environments.</div></div>","PeriodicalId":48686,"journal":{"name":"Sustainable Computing-Informatics & Systems","volume":"48 ","pages":"Article 101262"},"PeriodicalIF":5.7,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145614636","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

Dynamic task scheduling in cloud environments using improved K-means and DiffQ Neural Evolution approach 基于改进K-means和DiffQ神经进化方法的云环境下动态任务调度

IF 5.7 3区计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Sustainable Computing-Informatics & Systems

Pub Date : 2025-12-01 DOI: 10.1016/j.suscom.2025.101257

Divya R, Swapnil M. Parikh

Dynamic task scheduling in cloud computing for optimizing resource utilization and minimizing execution time, particularly in environments with fluctuating workloads and diverse application requirements. Conventional algorithms often struggle with scalability, computational complexity and slow convergence rates, leading to inefficiencies in resource management and increased costs. To address these challenges, this work proposes the Improved K-means EvoQ Framework which integrates enhanced k-means clustering with DiffQ Neural Evolution approach. This framework employs a global resource manager to monitor and optimize resource allocation while leveraging Differential Evolution (DE) and Deep Q-Learning (DQL) for dynamic policy optimization. The hybrid approach adapts to workload changes which enhances scalability and improves task scheduling efficiency by minimizing response time, waiting time and makespan while maximizing resource utilization. Comprehensive evaluation demonstrates the framework’s effectiveness across metrics such as Makespan, computation time, success rate and resource utilization making it a robust solution for dynamic task scheduling in cloud environments. Thus, this proposed framework provides a scalable, efficient and intelligent task scheduling solution paving the way for enhanced performance in modern cloud computing environments.

云计算中的动态任务调度，用于优化资源利用和最小化执行时间，特别是在工作负载波动和应用程序需求多样化的环境中。传统的算法往往与可扩展性、计算复杂性和缓慢的收敛速度作斗争，导致资源管理效率低下和成本增加。为了解决这些挑战，本工作提出了改进的K-means EvoQ框架，该框架将增强的K-means聚类与DiffQ神经进化方法相结合。该框架使用全局资源管理器来监控和优化资源分配，同时利用差分进化（DE）和深度q -学习（DQL）进行动态策略优化。混合方法适应工作负载变化，从而通过最小化响应时间、等待时间和完工时间来增强可伸缩性并提高任务调度效率，同时最大限度地提高资源利用率。综合评估证明了该框架在Makespan、计算时间、成功率和资源利用率等指标上的有效性，使其成为云环境中动态任务调度的健壮解决方案。因此，这个提议的框架提供了一个可伸缩、高效和智能的任务调度解决方案，为在现代云计算环境中增强性能铺平了道路。

{"title":"Dynamic task scheduling in cloud environments using improved K-means and DiffQ Neural Evolution approach","authors":"Divya R, Swapnil M. Parikh","doi":"10.1016/j.suscom.2025.101257","DOIUrl":"10.1016/j.suscom.2025.101257","url":null,"abstract":"<div><div>Dynamic task scheduling in cloud computing for optimizing resource utilization and minimizing execution time, particularly in environments with fluctuating workloads and diverse application requirements. Conventional algorithms often struggle with scalability, computational complexity and slow convergence rates, leading to inefficiencies in resource management and increased costs. To address these challenges, this work proposes the Improved K-means EvoQ Framework which integrates enhanced k-means clustering with DiffQ Neural Evolution approach. This framework employs a global resource manager to monitor and optimize resource allocation while leveraging Differential Evolution (DE) and Deep Q-Learning (DQL) for dynamic policy optimization. The hybrid approach adapts to workload changes which enhances scalability and improves task scheduling efficiency by minimizing response time, waiting time and makespan while maximizing resource utilization. Comprehensive evaluation demonstrates the framework’s effectiveness across metrics such as Makespan, computation time, success rate and resource utilization making it a robust solution for dynamic task scheduling in cloud environments. Thus, this proposed framework provides a scalable, efficient and intelligent task scheduling solution paving the way for enhanced performance in modern cloud computing environments.</div></div>","PeriodicalId":48686,"journal":{"name":"Sustainable Computing-Informatics & Systems","volume":"48 ","pages":"Article 101257"},"PeriodicalIF":5.7,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145614644","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

Energy consumption forecasting with hybrid deep learning approach, explainable AI, and hunger games optimization 混合深度学习方法、可解释人工智能和饥饿游戏优化的能源消耗预测

IF 5.7 3区计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Sustainable Computing-Informatics & Systems

Pub Date : 2025-12-01 DOI: 10.1016/j.suscom.2025.101255

Shahab S. Band , Faezeh Gholamrezaie , Fatemeh Asghari Hampa , Sultan Noman Qasem

Accurate forecasting of energy consumption is a critical component of effective resource management across the building, industrial, and transportation sectors. This work proposes a hybrid novel approach that incorporates Convolutional Neural Networks (CNN) with Gradient Boosting (GB) and Random Forest (FR) for improving energy demand prediction capabilities. These models will undergo an optimization process by the application of the Hunger Games Search (HGS) algorithm, boosting the prediction accuracy while incorporating Explainable AI (XAI) techniques that make the results interpretable.

In the PJM region, a regional transmission organization in the United States, the time series data recorded by four monitoring stations are considered. The performance of different models is evaluated based on critical metrics comprising Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), Mean Bias Error (MBE), and the coefficient of determination (R²) with 480 data points at each station. Among all models, CNN_RF_HGS performs the best as it tends to show a maximum of up to 0.9175 for the coefficient of determination in some cases. Such accuracy is achieved at the cost of a longer training time due to HGS optimization, highlighting a trade-off between accuracy and computational efficiency. However, the optimized model can be stored and reused as the pre-trained model, which will reduce the inference time by large margins and may fit real-time application purposes. Overall, this research demonstrates an effective blend of deep learning and traditional models for capturing complex nonlinear patterns in energy consumption, enabling more accurate and reliable forecasts.

准确预测能源消耗是建筑、工业和运输部门有效资源管理的关键组成部分。这项工作提出了一种混合的新方法，该方法将卷积神经网络（CNN）与梯度增强（GB）和随机森林（FR）相结合，以提高能源需求预测能力。这些模型将通过应用饥饿游戏搜索（HGS）算法进行优化，提高预测精度，同时结合可解释的人工智能（XAI）技术，使结果具有可解释性。在美国的一个区域传播组织PJM区域，考虑了四个监测站记录的时间序列数据。不同模型的性能基于关键指标，包括均方误差（MSE）、均方根误差（RMSE）、平均绝对百分比误差（MAPE）、平均偏差误差（MBE）和决定系数（R²），每个站点有480个数据点。在所有模型中，CNN_RF_HGS表现最好，在某些情况下，其决定系数的最大值可达0.9175。由于HGS优化，这种精度是以更长的训练时间为代价实现的，突出了精度和计算效率之间的权衡。然而，优化后的模型可以作为预训练模型存储和重用，这将大大减少推理时间，适合实时应用的目的。总的来说，这项研究证明了深度学习和传统模型的有效结合，可以捕获能源消耗中的复杂非线性模式，从而实现更准确和可靠的预测。

{"title":"Energy consumption forecasting with hybrid deep learning approach, explainable AI, and hunger games optimization","authors":"Shahab S. Band , Faezeh Gholamrezaie , Fatemeh Asghari Hampa , Sultan Noman Qasem","doi":"10.1016/j.suscom.2025.101255","DOIUrl":"10.1016/j.suscom.2025.101255","url":null,"abstract":"<div><div>Accurate forecasting of energy consumption is a critical component of effective resource management across the building, industrial, and transportation sectors. This work proposes a hybrid novel approach that incorporates Convolutional Neural Networks (CNN) with Gradient Boosting (GB) and Random Forest (FR) for improving energy demand prediction capabilities. These models will undergo an optimization process by the application of the Hunger Games Search (HGS) algorithm, boosting the prediction accuracy while incorporating Explainable AI (XAI) techniques that make the results interpretable.</div><div>In the PJM region, a regional transmission organization in the United States, the time series data recorded by four monitoring stations are considered. The performance of different models is evaluated based on critical metrics comprising Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), Mean Bias Error (MBE), and the coefficient of determination (R²) with 480 data points at each station. Among all models, CNN_RF_HGS performs the best as it tends to show a maximum of up to 0.9175 for the coefficient of determination in some cases. Such accuracy is achieved at the cost of a longer training time due to HGS optimization, highlighting a trade-off between accuracy and computational efficiency. However, the optimized model can be stored and reused as the pre-trained model, which will reduce the inference time by large margins and may fit real-time application purposes. Overall, this research demonstrates an effective blend of deep learning and traditional models for capturing complex nonlinear patterns in energy consumption, enabling more accurate and reliable forecasts.</div></div>","PeriodicalId":48686,"journal":{"name":"Sustainable Computing-Informatics & Systems","volume":"48 ","pages":"Article 101255"},"PeriodicalIF":5.7,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145614739","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

The application of bim technology in green building, design bim技术在绿色建筑设计中的应用

IF 5.7 3区计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Sustainable Computing-Informatics & Systems

Pub Date : 2025-12-01 DOI: 10.1016/j.suscom.2025.101244

Fang Liu

The construction sector faces growing pressure to reduce energy use and carbon emissions while meeting urban development needs. Green building practices have emerged as a response, yet their effectiveness is often limited by fragmented workflows and weak integration of sustainability principles. This study aims to evaluate BIM’s role in green building design by analyzing energy performance, resource allocation, and lifecycle sustainability metrics.

A case study methodology was adopted, focusing on certified green building projects with documented BIM adoption. Quantitative indicators—including energy saving rate, carbon footprint reduction, and operational cost savings—were used to assess performance. Results show that BIM-based design achieves 20–32 % energy savings, 15–22 % operational cost reductions, and 18–30 % decreases in carbon footprint compared to conventional approaches.

The findings confirm that BIM extends beyond design coordination to function as a comprehensive tool for sustainability evaluation. The study’s novelty lies in integrating BIM across the entire building lifecycle with measurable sustainability indices, offering practical insights for architects, engineers, and policymakers.

建筑行业在满足城市发展需求的同时，面临着越来越大的减少能源使用和碳排放的压力。绿色建筑实践作为一种回应而出现，但它们的有效性往往受到支离破碎的工作流程和可持续性原则整合不力的限制。本研究旨在通过分析能源性能、资源分配和生命周期可持续性指标来评估BIM在绿色建筑设计中的作用。采用了案例研究方法，重点关注认证的绿色建筑项目，并记录了BIM的采用情况。量化指标——包括节能率、碳足迹减少和运营成本节约——被用来评估绩效。结果表明，与传统方法相比，基于bim的设计可节省20 - 32% %的能源，降低15 - 22% %的运营成本，减少18 - 30% %的碳足迹。研究结果证实，BIM超越了设计协调，成为可持续性评估的综合工具。该研究的新颖之处在于将BIM与可衡量的可持续性指标整合到整个建筑生命周期中，为建筑师、工程师和政策制定者提供了实用的见解。

{"title":"The application of bim technology in green building, design","authors":"Fang Liu","doi":"10.1016/j.suscom.2025.101244","DOIUrl":"10.1016/j.suscom.2025.101244","url":null,"abstract":"<div><div>The construction sector faces growing pressure to reduce energy use and carbon emissions while meeting urban development needs. Green building practices have emerged as a response, yet their effectiveness is often limited by fragmented workflows and weak integration of sustainability principles. This study aims to evaluate BIM’s role in green building design by analyzing energy performance, resource allocation, and lifecycle sustainability metrics.</div><div>A case study methodology was adopted, focusing on certified green building projects with documented BIM adoption. Quantitative indicators—including energy saving rate, carbon footprint reduction, and operational cost savings—were used to assess performance. Results show that BIM-based design achieves 20–32 % energy savings, 15–22 % operational cost reductions, and 18–30 % decreases in carbon footprint compared to conventional approaches.</div><div>The findings confirm that BIM extends beyond design coordination to function as a comprehensive tool for sustainability evaluation. The study’s novelty lies in integrating BIM across the entire building lifecycle with measurable sustainability indices, offering practical insights for architects, engineers, and policymakers.</div></div>","PeriodicalId":48686,"journal":{"name":"Sustainable Computing-Informatics & Systems","volume":"48 ","pages":"Article 101244"},"PeriodicalIF":5.7,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145614642","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

HAPSO: An ACO-initialized, discretization-aware PSO for energy- and carbon-efficient VM consolidation in green cloud datacenters HAPSO：一种aco初始化、离散化感知的PSO，用于绿色云数据中心中节能和节能的VM整合

IF 5.7 3区计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Sustainable Computing-Informatics & Systems

Pub Date : 2025-12-01 DOI: 10.1016/j.suscom.2025.101258

Ali M. Baydoun , Ahmed S. Zekri

The energy demand of datacenters has been rising steadily, making them major contributors to global electricity consumption and carbon emissions. This paper proposes HAPSO, a hybrid metaheuristic that integrates Ant Colony Optimization (ACO) for initial virtual machine (VM) placement with discretization-aware Particle Swarm Optimization (PSO) for migration optimization, tailored for energy- and carbon-efficient VM consolidation in green cloud datacenters. In the first stage, ACO performs energy-aware placement of VMs onto physical hosts, emphasizing global search to satisfy resource constraints and minimize power usage. In the second stage, discrete PSO refines the allocation by migrating VMs from overloaded and underutilized hosts, focusing on local optimization to improve consolidation and reduce resource wastage. The novel contributions include: sequential metaheuristic hybridization, a system-informed particle initialization (seeding PSO with ACO output to ensure feasible starting solutions) and a heuristic-guided discretization method (mapping continuous updates into valid VM–host assignments), and a multi-objective fitness function that minimizes active servers and unused capacity to enhance efficiency. We implement HAPSO in CloudSimPlus and evaluate it on workloads ranging from 500 to 14,000 VMs using realistic trace-driven simulations. Results show that HAPSO reduces energy consumption by 6.72 % on average (up to 10.0 %) and carbon emissions by 10.5 %, with savings peaking at 25.8 % in mid-scale workloads, compared to the ACO baseline, while maintaining SLA compliance. Statistical significance is confirmed via Friedman, Kendall’s W, and Wilcoxon signed-rank tests, with large effect sizes. These findings highlight HAPSO’s potential to support greener, sustainable cloud operations.

数据中心的能源需求一直在稳步增长，使其成为全球电力消耗和碳排放的主要贡献者。本文提出了一种混合元启发式算法HAPSO，该算法集成了用于初始虚拟机（VM）放置的蚁群优化（ACO）和用于迁移优化的离散化感知粒子群优化（PSO），专为绿色云数据中心中节能和节能的VM整合而设计。在第一阶段，蚁群算法在物理主机上执行能量感知的虚拟机布局，强调全局搜索以满足资源约束并最小化功耗。在第二阶段，离散PSO通过从过载和未充分利用的主机迁移虚拟机来优化分配，专注于局部优化以提高整合并减少资源浪费。新的贡献包括：顺序元启发式杂交，系统通知粒子初始化（用蚁群算法输出播种PSO以确保可行的起始解）和启发式引导离散化方法（将连续更新映射到有效的vm -主机分配），以及多目标适应度函数，该函数最小化活动服务器和未使用容量以提高效率。我们在CloudSimPlus中实现了HAPSO，并使用真实的跟踪驱动模拟在500到14,000 vm的工作负载上对其进行了评估。结果表明，与ACO基线相比，HAPSO平均降低了6.72 %（最高可达10.0 %），碳排放量降低了10. %，在中等规模的工作负载中，节约的峰值为25.8 %，同时保持了SLA合规性。统计显著性通过Friedman， Kendall 's W和Wilcoxon sign -rank检验证实，具有较大的效应量。这些发现凸显了HAPSO在支持更环保、可持续的云运营方面的潜力。

{"title":"HAPSO: An ACO-initialized, discretization-aware PSO for energy- and carbon-efficient VM consolidation in green cloud datacenters","authors":"Ali M. Baydoun , Ahmed S. Zekri","doi":"10.1016/j.suscom.2025.101258","DOIUrl":"10.1016/j.suscom.2025.101258","url":null,"abstract":"<div><div>The energy demand of datacenters has been rising steadily, making them major contributors to global electricity consumption and carbon emissions. This paper proposes HAPSO, a hybrid metaheuristic that integrates Ant Colony Optimization (ACO) for initial virtual machine (VM) placement with discretization-aware Particle Swarm Optimization (PSO) for migration optimization, tailored for energy- and carbon-efficient VM consolidation in green cloud datacenters. In the first stage, ACO performs energy-aware placement of VMs onto physical hosts, emphasizing global search to satisfy resource constraints and minimize power usage. In the second stage, discrete PSO refines the allocation by migrating VMs from overloaded and underutilized hosts, focusing on local optimization to improve consolidation and reduce resource wastage. The novel contributions include: sequential metaheuristic hybridization, a system-informed particle initialization (seeding PSO with ACO output to ensure feasible starting solutions) and a heuristic-guided discretization method (mapping continuous updates into valid VM–host assignments), and a multi-objective fitness function that minimizes active servers and unused capacity to enhance efficiency. We implement HAPSO in CloudSimPlus and evaluate it on workloads ranging from 500 to 14,000 VMs using realistic trace-driven simulations. Results show that HAPSO reduces energy consumption by 6.72 % on average (up to 10.0 %) and carbon emissions by 10.5 %, with savings peaking at 25.8 % in mid-scale workloads, compared to the ACO baseline, while maintaining SLA compliance. Statistical significance is confirmed via Friedman, Kendall’s W, and Wilcoxon signed-rank tests, with large effect sizes. These findings highlight HAPSO’s potential to support greener, sustainable cloud operations.</div></div>","PeriodicalId":48686,"journal":{"name":"Sustainable Computing-Informatics & Systems","volume":"48 ","pages":"Article 101258"},"PeriodicalIF":5.7,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145614738","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

Economic and environmental optimization-dispatch in large-scale power systems using weighted mean of vectors algorithm 基于向量加权平均算法的大型电力系统经济与环境优化调度

IF 5.7 3区计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Sustainable Computing-Informatics & Systems

Pub Date : 2025-12-01 DOI: 10.1016/j.suscom.2025.101265

Abdullah M. Shaheen , Ali M. El-Rifaie , Badr Al Faiya , Ghareeb Moustafa , Hashim Alnami

Large-scale optimization in Combined Economic and Environmental dispatch (CEED) is crucial for improving electrical power system management. This study introduces a developed weIghted meaN oF vectOrs Technique (INFOT) algorithm tailored for the CEED problem, featuring three primary operators, vector combining, rule updating, and local searching, that collaboratively optimizes generation costs and reduces environmental emissions. Addi The developed INFOT algorithm is utilized to solve the CEED problem and tested on two large scale power system with 40 and 160 thermal units. The INFOT algorithm is compared with several recent optimization techniques. For the 40-unit power generation system with a load demand of 10,500 MW, the proposed INFOT algorithm achieves a 5.6 % reduction in total fuel costs in cost minimization (Scenario 1) compared to the best competitor, while showing a significant improvement in emissions. Specifically, INFOT reduces emissions from 386,946 kg/h in Scenario 1–200,138.8 kg/h in costs and emissions minimization (Scenario 2), representing a 48.3 % reduction. Additionally, the generator output analysis indicates that INFOT can balance the generation requirements, preventing excessive stress on any particular unit and improving overall system stability. The study confirms that INFOT is a competitive and reliable optimization method for addressing CEED problems, effectively managing load variations and generator outputs over a 24-hour period. To validate its practical applicability, the proposed INFOT algorithm was applied to the IEEE 30-bus system for emission minimization. Comparative results demonstrate INFOT’s superior convergence speed and lowest emission levels relative to several state-of-the-art algorithms.

经济与环境联合调度中的大规模优化是提高电力系统管理水平的关键。本文介绍了一种针对CEED问题开发的加权向量均值技术（INFOT）算法，该算法具有三个主要操作，即向量组合、规则更新和局部搜索，可协同优化发电成本并减少环境排放。将所提出的INFOT算法应用于求解CEED问题，并在40和160个热电机组的大型电力系统上进行了测试。将INFOT算法与最近的几种优化技术进行了比较。对于负载需求为10,500 MW的40台发电系统，与最佳竞争对手相比，所提出的INFOT算法在成本最小化（场景1）下实现了5.6% %的总燃料成本降低，同时显示出显著的排放改善。具体来说，在情景1中，INFOT的排放量减少了386,946 kg/h，成本和排放量减少了200,138.8 kg/h（情景2），减少了48.3% %。此外，发电机输出分析表明，INFOT可以平衡发电需求，防止任何特定机组承受过大压力，提高系统整体稳定性。该研究证实，INFOT是解决CEED问题的一种有竞争力且可靠的优化方法，可以有效地管理24小时内的负载变化和发电机输出。为了验证该算法的实用性，将该算法应用于IEEE 30总线系统，实现了排放最小化。对比结果表明，相对于几种最先进的算法，INFOT具有优越的收敛速度和最低的排放水平。

{"title":"Economic and environmental optimization-dispatch in large-scale power systems using weighted mean of vectors algorithm","authors":"Abdullah M. Shaheen , Ali M. El-Rifaie , Badr Al Faiya , Ghareeb Moustafa , Hashim Alnami","doi":"10.1016/j.suscom.2025.101265","DOIUrl":"10.1016/j.suscom.2025.101265","url":null,"abstract":"<div><div>Large-scale optimization in Combined Economic and Environmental dispatch (CEED) is crucial for improving electrical power system management. This study introduces a developed weIghted meaN oF vectOrs Technique (INFOT) algorithm tailored for the CEED problem, featuring three primary operators, vector combining, rule updating, and local searching, that collaboratively optimizes generation costs and reduces environmental emissions. Addi The developed INFOT algorithm is utilized to solve the CEED problem and tested on two large scale power system with 40 and 160 thermal units. The INFOT algorithm is compared with several recent optimization techniques. For the 40-unit power generation system with a load demand of 10,500 MW, the proposed INFOT algorithm achieves a 5.6 % reduction in total fuel costs in cost minimization (Scenario 1) compared to the best competitor, while showing a significant improvement in emissions. Specifically, INFOT reduces emissions from 386,946 kg/h in Scenario 1–200,138.8 kg/h in costs and emissions minimization (Scenario 2), representing a 48.3 % reduction. Additionally, the generator output analysis indicates that INFOT can balance the generation requirements, preventing excessive stress on any particular unit and improving overall system stability. The study confirms that INFOT is a competitive and reliable optimization method for addressing CEED problems, effectively managing load variations and generator outputs over a 24-hour period. To validate its practical applicability, the proposed INFOT algorithm was applied to the IEEE 30-bus system for emission minimization. Comparative results demonstrate INFOT’s superior convergence speed and lowest emission levels relative to several state-of-the-art algorithms.</div></div>","PeriodicalId":48686,"journal":{"name":"Sustainable Computing-Informatics & Systems","volume":"48 ","pages":"Article 101265"},"PeriodicalIF":5.7,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145614645","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

Energy efficient optimization of renewable energy dispatch using blockchain-verified deep reinforcement learning controllers 利用区块链验证的深度强化学习控制器对可再生能源调度进行节能优化

IF 5.7 3区计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Sustainable Computing-Informatics & Systems

Pub Date : 2025-12-01 DOI: 10.1016/j.suscom.2025.101256

Murugan Marimuthu , Padmaja Kadiri , Senthilkumar Ganapathy , Venkatesh Kumar Pandiyan

The intelligent control issue of renewable energy dispatching in micro grids relates to the scenario of energy efficiency and transaction security. An innovative framework is introduced on the real-time scheduling of solar and wind energy over a distributed network that includes DRL-based controllers embedded in a blockchain-authenticated dispatch-protocol. DRA PPO is applied by the DRA agent to optimize strategies of power distribution dynamically across multiple prosumer nodes under the influence of stochastic generation and consumption profiles. The blockchain layer is specifically made to validate dispatch decision with the help of smart contracts, which guarantee integrity of data, tamper-proof scheduling, and transparent peer-to-peer energy exchange. An OPAL-RT + Hyperledger Fabric testbed was experimentally validated to 96.8 Renewable Dispatch Accuracy, 19.5 Energy Loss Reduction and 14.3 Grid Stability Improvement and transaction finality is within 2.1 s. Economic analysis also denoted a 25 % cost saving by prosumers relative to rule-based control. This decentralized control architecture has therefore been proven to be scalable to heterogeneous groups of microgrids, resilient to node failure, or cyber-attacks. Combination of DAR with blockchain creates a safe, self-reinforcing, and energy efficient attention framework perfectly fit in the coming generation of green energy dispatch frameworks.

微电网可再生能源调度的智能控制问题涉及到能源效率和交易安全的场景。在分布式网络上引入了太阳能和风能实时调度的创新框架，该网络包括嵌入在区块链认证调度协议中的基于drl的控制器。DRA - PPO是DRA agent在随机发电和随机消费情况下，应用于多产消节点间动态优化电力分配策略的一种方法。区块链层是专门用来通过智能合约验证调度决策的，它保证了数据的完整性、防篡改调度和透明的点对点能源交换。OPAL-RT + Hyperledger Fabric测试平台实验验证了96.8的可再生调度精度，19.5的能量损失减少和14.3的电网稳定性提高，交易最终性在2.1 s以内。经济分析还表明，相对于基于规则的控制，生产消费者节省了25%的成本。因此，这种分散的控制体系结构已被证明可扩展到异构微电网组，对节点故障或网络攻击具有弹性。DAR与区块链的结合创造了一个安全、自我强化、节能的关注框架，非常适合下一代绿色能源调度框架。

{"title":"Energy efficient optimization of renewable energy dispatch using blockchain-verified deep reinforcement learning controllers","authors":"Murugan Marimuthu , Padmaja Kadiri , Senthilkumar Ganapathy , Venkatesh Kumar Pandiyan","doi":"10.1016/j.suscom.2025.101256","DOIUrl":"10.1016/j.suscom.2025.101256","url":null,"abstract":"<div><div>The intelligent control issue of renewable energy dispatching in micro grids relates to the scenario of energy efficiency and transaction security. An innovative framework is introduced on the real-time scheduling of solar and wind energy over a distributed network that includes DRL-based controllers embedded in a blockchain-authenticated dispatch-protocol. DRA PPO is applied by the DRA agent to optimize strategies of power distribution dynamically across multiple prosumer nodes under the influence of stochastic generation and consumption profiles. The blockchain layer is specifically made to validate dispatch decision with the help of smart contracts, which guarantee integrity of data, tamper-proof scheduling, and transparent peer-to-peer energy exchange. An OPAL-RT + Hyperledger Fabric testbed was experimentally validated to 96.8 Renewable Dispatch Accuracy, 19.5 Energy Loss Reduction and 14.3 Grid Stability Improvement and transaction finality is within 2.1 s. Economic analysis also denoted a 25 % cost saving by prosumers relative to rule-based control. This decentralized control architecture has therefore been proven to be scalable to heterogeneous groups of microgrids, resilient to node failure, or cyber-attacks. Combination of DAR with blockchain creates a safe, self-reinforcing, and energy efficient attention framework perfectly fit in the coming generation of green energy dispatch frameworks.</div></div>","PeriodicalId":48686,"journal":{"name":"Sustainable Computing-Informatics & Systems","volume":"48 ","pages":"Article 101256"},"PeriodicalIF":5.7,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145684123","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 hybrid fuzzy logic and deep reinforcement learning algorithm for adaptive task scheduling and resource allocation in heterogeneous Fog–Cloud environments 一种用于异构雾云环境下自适应任务调度和资源分配的混合模糊逻辑和深度强化学习算法

IF 5.7 3区计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Sustainable Computing-Informatics & Systems

Pub Date : 2025-11-27 DOI: 10.1016/j.suscom.2025.101260

Setareh Moazzami , Abbas Mirzaei , Mehdi Aminian , Ramin Karimi , Nasser Mikaeilvand

Intelligent task scheduling in distributed computing environments such as Fog–Cloud systems remains a significant challenge, particularly in the context of the Internet of Things (IoT), where multiple objectives such as minimizing delay, energy consumption, and makespan must be simultaneously addressed. This paper proposes an adaptive hybrid framework that integrates fuzzy logic with Deep Q-Network (DQN) reinforcement learning to optimize task scheduling and resource allocation in heterogeneous and dynamic environments. The model is designed to maintain service quality while remaining compatible with limited computational resources. The scheduling problem is first formulated as a multi-objective optimization model aimed at jointly minimizing delay, energy usage, and makespan. A fuzzy inference system is then employed to evaluate task attributes such as deadline, delay sensitivity, and data volume in order to assign priority levels. Based on this prioritization, the DQN agent dynamically allocates resources by interacting with the environment and learning from feedback. The proposed framework was evaluated on scenarios involving 500–2000 tasks under varying resource conditions, and its performance was benchmarked against conventional algorithms. Experimental results demonstrate that the proposed method achieves, on average, a 27.8 % reduction in execution time, a 29.6 % decrease in scheduling delay, an 18 % reduction in energy consumption, and a 21.4 % improvement in makespan. These outcomes highlight the framework’s effectiveness in balancing accuracy, responsiveness, and resource efficiency, making it well-suited for deployment in real-world, heterogeneous, and dynamically loaded computing environments.

在分布式计算环境（如Fog-Cloud系统）中，智能任务调度仍然是一个重大挑战，特别是在物联网（IoT）的背景下，必须同时解决多个目标，如最小化延迟、能耗和完工时间。本文提出了一种将模糊逻辑与Deep Q-Network （DQN）强化学习相结合的自适应混合框架，以优化异构和动态环境下的任务调度和资源分配。该模型旨在保持服务质量，同时与有限的计算资源保持兼容。首先将调度问题表述为以延误、能耗和完工时间共同最小化为目标的多目标优化模型。然后使用模糊推理系统来评估任务属性，如截止日期、延迟敏感性和数据量，以分配优先级级别。基于这种优先级，DQN代理通过与环境交互和从反馈中学习来动态分配资源。在不同资源条件下，对涉及500-2000个 任务的场景进行了评估，并将其性能与传统算法进行了基准测试。实验结果表明，该方法的平均执行时间减少27.8% %，调度延迟减少29.6 %，能耗减少18 %，完工时间提高21.4% %。这些结果突出了框架在平衡准确性、响应性和资源效率方面的有效性，使其非常适合在真实的、异构的和动态加载的计算环境中部署。

{"title":"A hybrid fuzzy logic and deep reinforcement learning algorithm for adaptive task scheduling and resource allocation in heterogeneous Fog–Cloud environments","authors":"Setareh Moazzami , Abbas Mirzaei , Mehdi Aminian , Ramin Karimi , Nasser Mikaeilvand","doi":"10.1016/j.suscom.2025.101260","DOIUrl":"10.1016/j.suscom.2025.101260","url":null,"abstract":"<div><div>Intelligent task scheduling in distributed computing environments such as Fog–Cloud systems remains a significant challenge, particularly in the context of the Internet of Things (IoT), where multiple objectives such as minimizing delay, energy consumption, and makespan must be simultaneously addressed. This paper proposes an adaptive hybrid framework that integrates fuzzy logic with Deep Q-Network (DQN) reinforcement learning to optimize task scheduling and resource allocation in heterogeneous and dynamic environments. The model is designed to maintain service quality while remaining compatible with limited computational resources. The scheduling problem is first formulated as a multi-objective optimization model aimed at jointly minimizing delay, energy usage, and makespan. A fuzzy inference system is then employed to evaluate task attributes such as deadline, delay sensitivity, and data volume in order to assign priority levels. Based on this prioritization, the DQN agent dynamically allocates resources by interacting with the environment and learning from feedback. The proposed framework was evaluated on scenarios involving 500–2000 tasks under varying resource conditions, and its performance was benchmarked against conventional algorithms. Experimental results demonstrate that the proposed method achieves, on average, a 27.8 % reduction in execution time, a 29.6 % decrease in scheduling delay, an 18 % reduction in energy consumption, and a 21.4 % improvement in makespan. These outcomes highlight the framework’s effectiveness in balancing accuracy, responsiveness, and resource efficiency, making it well-suited for deployment in real-world, heterogeneous, and dynamically loaded computing environments.</div></div>","PeriodicalId":48686,"journal":{"name":"Sustainable Computing-Informatics & Systems","volume":"49 ","pages":"Article 101260"},"PeriodicalIF":5.7,"publicationDate":"2025-11-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145694673","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

The implementation of machine learning models for predicting CO2 emission in carbon capture and storage systems 机器学习模型在碳捕获和储存系统中预测二氧化碳排放的实现

IF 5.7 3区计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Sustainable Computing-Informatics & Systems

Pub Date : 2025-11-26 DOI: 10.1016/j.suscom.2025.101264

Zixiang Xu , Xiaokai Zhou , Yishan Wang

Although Carbon Capture and Storage (CCS) systems are essential for lowering CO₂ emissions worldwide, their operational complexity and fluctuating performance factors make it difficult to predict emission levels within these systems. This issue is significant since it directly affects maximizing CCS efficiency, lowering environmental risks, and directing investment and policy choices related to climate mitigation. This research uses six machine learning (ML) models, Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), Gradient Boosting (GB), Extreme Gradient Boosting (XGBoost), and Categorical Boosting (CatBoost), all of which are optimized employing GridSearchCV and the Hybrid Kepler Optimization Algorithm (HKOA) to solve this issue and provide a thorough framework for the prediction of CO₂ emissions within CCS systems. According to the robust metrics’ results, the HKOA-optimized XGBoost model outperformed all others, achieving the lowest error rates (MAE = 22.6 train, 47.5 test; RMSE = 42.1 train, 196.0 test), R² values of 0.9994 (training) and 0.9879 (testing), and an objective function (OBJ) value of 15385.2 in testing. Moreover, it achieved a high A20 accuracy (≥72 %) and low uncertainty index (U95), underscoring its robust generalization and minimal deviation under uncertainty. "Gas" and "activity" were identified as important predictors by SHAP and feature importance analyses. These results not only show how well tree-based ensemble models predict emissions, but they also provide useful tools for practical use in CCS planning, monitoring, and optimization, which helps create more sustainable and profitable carbon management plans.

尽管碳捕集与封存（CCS）系统对于降低全球二氧化碳排放至关重要，但其操作复杂性和波动的性能因素使得难以预测这些系统的排放水平。这个问题很重要，因为它直接影响到最大限度地提高CCS效率，降低环境风险，以及指导与减缓气候变化有关的投资和政策选择。本研究采用了逻辑回归（LR）、决策树（DT）、随机森林（RF）、梯度增强（GB）、极限梯度增强（XGBoost）和分类增强（CatBoost） 6种机器学习（ML）模型，采用GridSearchCV和混合开普勒优化算法（HKOA）对这些模型进行优化，解决了这一问题，并为CCS系统内二氧化碳排放的预测提供了一个全面的框架。从鲁棒性指标结果来看，hkoa优化的XGBoost模型优于其他模型，错误率最低（MAE = 22.6 train, 47.5 test; RMSE = 42.1 train, 196.0 test）， R2值为0.9994 （training）和0.9879 (testing)，测试目标函数（OBJ）值为15385.2。A20精度高（≥72 %），不确定度指数低（U95），泛化能力强，在不确定条件下偏差最小。通过SHAP和特征重要性分析，确定“气”和“活度”是重要的预测因子。这些结果不仅显示了基于树木的集合模型预测排放的效果，而且还为CCS规划、监测和优化提供了实用的工具，有助于制定更可持续、更有利可图的碳管理计划。

{"title":"The implementation of machine learning models for predicting CO2 emission in carbon capture and storage systems","authors":"Zixiang Xu , Xiaokai Zhou , Yishan Wang","doi":"10.1016/j.suscom.2025.101264","DOIUrl":"10.1016/j.suscom.2025.101264","url":null,"abstract":"<div><div>Although Carbon Capture and Storage (CCS) systems are essential for lowering CO<sub>2</sub> emissions worldwide, their operational complexity and fluctuating performance factors make it difficult to predict emission levels within these systems. This issue is significant since it directly affects maximizing CCS efficiency, lowering environmental risks, and directing investment and policy choices related to climate mitigation. This research uses six machine learning (ML) models, Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), Gradient Boosting (GB), Extreme Gradient Boosting (XGBoost), and Categorical Boosting (CatBoost), all of which are optimized employing GridSearchCV and the Hybrid Kepler Optimization Algorithm (HKOA) to solve this issue and provide a thorough framework for the prediction of CO₂ emissions within CCS systems. According to the robust metrics’ results, the HKOA-optimized XGBoost model outperformed all others, achieving the lowest error rates (MAE = 22.6 train, 47.5 test; RMSE = 42.1 train, 196.0 test), R<sup>2</sup> values of 0.9994 (training) and 0.9879 (testing), and an objective function (OBJ) value of 15385.2 in testing. Moreover, it achieved a high A20 accuracy (≥72 %) and low uncertainty index (U95), underscoring its robust generalization and minimal deviation under uncertainty. \"Gas\" and \"activity\" were identified as important predictors by SHAP and feature importance analyses. These results not only show how well tree-based ensemble models predict emissions, but they also provide useful tools for practical use in CCS planning, monitoring, and optimization, which helps create more sustainable and profitable carbon management plans.</div></div>","PeriodicalId":48686,"journal":{"name":"Sustainable Computing-Informatics & Systems","volume":"49 ","pages":"Article 101264"},"PeriodicalIF":5.7,"publicationDate":"2025-11-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145749254","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

An energy-efficient deep learning model evaluation for robust image recognition in automated decision-making systems 一种用于自动决策系统鲁棒图像识别的节能深度学习模型评估

IF 5.7 3区计算机科学 Q1 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Sustainable Computing-Informatics & Systems

Pub Date : 2025-11-13 DOI: 10.1016/j.suscom.2025.101254

Chen Tao

Accurate image recognition and classification in automated decision-making systems needs a significant deep learning model with an ability of managing the large volumetric Data. The traditional convolutional model is often fails to captures the spatial dependencies in the image, that limiting the accuracy of the model in several domains. Convolutional neural networks (CNNs) characterize a period of deep learning processes frequently applied in computer vision, which may be used to examine images and assign learnable weights to distinct objects in the image. This study compares the 3D V-Net, YOLOv4-EfficientNet, Grad-CAM, Gabor CNN, and Deep Feedforward Network deep learning model advancements and evaluates the most reliable model for robust image recognition in decision-making systems in various domains. The concert of each model is tested by means of performance metrics that includes, Precision, F1 Score, Recall, Accuracy, Intersection over Union (IoU), Dice Coefficient, Mean Squared Error (MSE), mean Average Precision (mAP) and Mean Absolute Error (MAE). Comparative analysis showcases that, the 3D V-Net mode surpasses the other models in by achieving the higher IoU of 85.4 % and Dice Coefficient of 90.3 %) whereas the Gabor CNN balances accuracy and the computational efficiency. The YOLOv4-EfficientNet and Grad-CAM offers a transparency in classification decisions. The results showcase that, the selected model is determined by application demands, with the 3D V-Net remains a most significant for image recognition and automated decision-making systems.

在自动决策系统中，准确的图像识别和分类需要一个重要的深度学习模型，该模型具有管理大容量数据的能力。传统的卷积模型往往不能捕捉图像中的空间依赖关系，这限制了模型在多个领域的准确性。卷积神经网络（cnn）表征了一段深度学习过程，它经常应用于计算机视觉中，可用于检查图像并为图像中的不同对象分配可学习的权重。本研究比较了3D V-Net、yolov4 - effentnet、Grad-CAM、Gabor CNN和Deep Feedforward Network深度学习模型的进步，并评估了在各个领域决策系统中鲁棒图像识别的最可靠模型。每个模型的一致性通过性能指标进行测试，这些指标包括精度，F1分数，召回率，准确性，交集超过联盟（IoU），骰子系数，均方误差（MSE），平均平均精度（mAP）和平均绝对误差（MAE）。对比分析表明，3D V-Net模式在IoU（85.4 %）和Dice系数（90.3 %）方面优于其他模型，而Gabor CNN在精度和计算效率方面取得了平衡。YOLOv4-EfficientNet和Grad-CAM为分类决策提供了透明度。结果表明，所选择的模型是由应用需求决定的，其中3D V-Net在图像识别和自动决策系统中仍然是最重要的。

{"title":"An energy-efficient deep learning model evaluation for robust image recognition in automated decision-making systems","authors":"Chen Tao","doi":"10.1016/j.suscom.2025.101254","DOIUrl":"10.1016/j.suscom.2025.101254","url":null,"abstract":"<div><div>Accurate image recognition and classification in automated decision-making systems needs a significant deep learning model with an ability of managing the large volumetric Data. The traditional convolutional model is often fails to captures the spatial dependencies in the image, that limiting the accuracy of the model in several domains. Convolutional neural networks (CNNs) characterize a period of deep learning processes frequently applied in computer vision, which may be used to examine images and assign learnable weights to distinct objects in the image. This study compares the 3D V-Net, YOLOv4-EfficientNet, Grad-CAM, Gabor CNN, and Deep Feedforward Network deep learning model advancements and evaluates the most reliable model for robust image recognition in decision-making systems in various domains. The concert of each model is tested by means of performance metrics that includes, Precision, F1 Score, Recall, Accuracy, Intersection over Union (IoU), Dice Coefficient, Mean Squared Error (MSE), mean Average Precision (mAP) and Mean Absolute Error (MAE). Comparative analysis showcases that, the 3D V-Net mode surpasses the other models in by achieving the higher IoU of 85.4 % and Dice Coefficient of 90.3 %) whereas the Gabor CNN balances accuracy and the computational efficiency. The YOLOv4-EfficientNet and Grad-CAM offers a transparency in classification decisions. The results showcase that, the selected model is determined by application demands, with the 3D V-Net remains a most significant for image recognition and automated decision-making systems.</div></div>","PeriodicalId":48686,"journal":{"name":"Sustainable Computing-Informatics & Systems","volume":"48 ","pages":"Article 101254"},"PeriodicalIF":5.7,"publicationDate":"2025-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145568408","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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