Engineering Applications of Artificial Intelligence最新文献_第6页

Identification of acoustic emission wire breakage signals in bridge cables under imbalanced data conditions 数据不平衡条件下桥梁电缆声发射断线信号的识别

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-02-10 DOI: 10.1016/j.engappai.2026.114085

Kaixuan Hui , Guangming Li , Guizhen Niu , Shuai Zhao

Deep learning techniques have demonstrated significant potential for the quantitative identification of Acoustic Emission (AE) wire breakage signals generated by wire breakage in bridge cables, which is critical for ensuring the structural safety of cable-supported bridges. However, under actual bridge operating conditions, AE wire breakage signals are extremely limited and difficult to collect. This leads to data imbalance due to scarcity of wire breakage samples and significantly reduces the accuracy of wire breakage identification. To address this challenge, we propose an innovative method combining a Wasserstein distance and Gradient penalty-enhanced Deep Convolutional Generative Adversarial Network (WGDCGAN) and the Shifted windows (Swin) Transformer model for AE wire breakage signal identification. The effectiveness of the proposed method is validated through full-scale bridge cable tests. Experimental results demonstrate that the proposed approach achieves superior performance in terms of total accuracy A_T, breakage true detection rate B_T (equivalent to sensitivity), and F₁-Score F₁, while effectively overcoming the performance degradation typically caused by imbalanced data. These findings highlight the method's strong potential for improving the reliability of AE-based wire breakage monitoring in bridge engineering.

深度学习技术在定量识别桥梁电缆断线产生的声发射断线信号方面已经显示出巨大的潜力，这对于确保索桥的结构安全至关重要。然而，在实际桥梁运行条件下，声发射断线信号极为有限，难以采集。由于断丝样本的稀缺，导致数据不平衡，大大降低了断丝鉴定的准确性。为了解决这一挑战，我们提出了一种结合Wasserstein距离和梯度惩罚增强的深度卷积生成对抗网络（WGDCGAN）和移位窗口（Swin）变压器模型的创新方法，用于声发射断线信号识别。通过全尺寸桥索试验验证了该方法的有效性。实验结果表明，该方法在总准确率AT、破损真检测率BT（相当于灵敏度）和F1- score F1方面均取得了较好的性能，同时有效克服了数据不平衡通常导致的性能下降。这些发现凸显了该方法在提高桥梁工程中基于ae的导线断裂监测可靠性方面的巨大潜力。

{"title":"Identification of acoustic emission wire breakage signals in bridge cables under imbalanced data conditions","authors":"Kaixuan Hui , Guangming Li , Guizhen Niu , Shuai Zhao","doi":"10.1016/j.engappai.2026.114085","DOIUrl":"10.1016/j.engappai.2026.114085","url":null,"abstract":"<div><div>Deep learning techniques have demonstrated significant potential for the quantitative identification of Acoustic Emission (AE) wire breakage signals generated by wire breakage in bridge cables, which is critical for ensuring the structural safety of cable-supported bridges. However, under actual bridge operating conditions, AE wire breakage signals are extremely limited and difficult to collect. This leads to data imbalance due to scarcity of wire breakage samples and significantly reduces the accuracy of wire breakage identification. To address this challenge, we propose an innovative method combining a Wasserstein distance and Gradient penalty-enhanced Deep Convolutional Generative Adversarial Network (WGDCGAN) and the Shifted windows (Swin) Transformer model for AE wire breakage signal identification. The effectiveness of the proposed method is validated through full-scale bridge cable tests. Experimental results demonstrate that the proposed approach achieves superior performance in terms of total accuracy <em>A</em><sub>T</sub>, breakage true detection rate <em>B</em><sub>T</sub> (equivalent to sensitivity), and <em>F</em><sub>1</sub>-Score <em>F</em><sub>1</sub>, while effectively overcoming the performance degradation typically caused by imbalanced data. These findings highlight the method's strong potential for improving the reliability of AE-based wire breakage monitoring in bridge engineering.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 114085"},"PeriodicalIF":8.0,"publicationDate":"2026-02-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175222","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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

Enhanced recognition of low-discernibility Railway Sleeper Serial Numbers via dual-stage adaptive image enhancement and position prior-guided detection 基于双级自适应图像增强和位置先验引导检测的铁路轨枕序列号识别方法

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-02-10 DOI: 10.1016/j.engappai.2026.114110

Peng Shi , Jingjing Guo , Lu Deng , Yingkai Liu , Lizhi Long , Shaopeng Xu

Automatic recognition of Railway Sleeper Serial Numbers (RSSNs) is essential for traceability, quality management, and lifecycle maintenance of railway infrastructure. In practice, embossed serial numbers on concrete surfaces exhibit extremely low discernibility with minimal height variations (<1 mm) and negligible color differentiation. Traditional Red-Green-Blue-based (RGB-based) image enhancement and Optical Character Recognition (OCR) methods face a fundamental limitation: they cannot directly capture the three-dimensional geometric features distinguishing embossed characters from their surroundings. To address this challenge, this study proposes an integrated framework based on line laser height imaging and position prior-guided detection with three key innovations: (1) a cascaded processing framework leverages geometric height information to overcome RGB-based method limitations; (2) a Dual-stage Adaptive Image Enhancement (DAIE) strategy converts 16 binary-digit (bit) height images into optimized 8-bit visualizations by systematically selecting optimal methods: modulo truncation for global structure and Minimum-Maximum (Min-Max) normalization for local detail enhancement; and (3) a Position Prior-guided Spatial Attention (PPSA) Feature Pyramid Network (FPN) integrates statistically-derived position priors to enhance small target detection. Comprehensive validation on 2234 images demonstrates superior performance: 98.2% F1-score and 99.38% recognition accuracy at 27 Frames Per Second (FPS), achieving 2.4% improvement over state-of-the-art methods. Ablation experiments confirm the individual contributions of the PPSA module (4.0%), the Small Target Enhancement (STE) module (1.4%), and the DAIE strategy (3.08%). Field testing in a prefabricated factory validates industrial applicability, providing a scalable technical framework and valuable reference for low-discernibility embossed industrial character recognition. Code is publicly available at https://github.com/shipeng38/RSSN-recognition.

铁路轨枕序列号的自动识别对铁路基础设施的可追溯性、质量管理和生命周期维护至关重要。在实践中，混凝土表面上的浮雕序列号表现出极低的可辨性，高度变化最小（1毫米），颜色差异可以忽略不计。传统的基于红-绿-蓝（rgb）的图像增强和光学字符识别（OCR）方法面临着一个根本性的局限性：它们不能直接捕获将浮雕字符与周围环境区分开来的三维几何特征。为了解决这一挑战，本研究提出了一个基于线激光高度成像和位置先验制导检测的集成框架，其中有三个关键创新：(1)级联处理框架利用几何高度信息克服基于rgb方法的局限性；(2)双阶段自适应图像增强（DAIE）策略通过系统选择最优方法将16位二进制（bit）高度图像转换为优化的8位可视化图像：模截断用于全局结构，最小-最大（Min-Max）归一化用于局部细节增强；(3)位置先验引导的空间注意（PPSA）特征金字塔网络（FPN）集成了统计导出的位置先验，增强了小目标检测。对2234张图像的综合验证显示了卓越的性能：在27帧每秒（FPS）下，f1得分为98.2%，识别准确率为99.38%，比最先进的方法提高了2.4%。烧蚀实验证实了PPSA模块（4.0%）、小目标增强（STE）模块（1.4%）和DAIE策略（3.08%）的个人贡献。在预制工厂的现场测试验证了工业适用性，为低可辨度压花工业字符识别提供了可扩展的技术框架和有价值的参考。代码可在https://github.com/shipeng38/RSSN-recognition上公开获取。

{"title":"Enhanced recognition of low-discernibility Railway Sleeper Serial Numbers via dual-stage adaptive image enhancement and position prior-guided detection","authors":"Peng Shi , Jingjing Guo , Lu Deng , Yingkai Liu , Lizhi Long , Shaopeng Xu","doi":"10.1016/j.engappai.2026.114110","DOIUrl":"10.1016/j.engappai.2026.114110","url":null,"abstract":"<div><div>Automatic recognition of Railway Sleeper Serial Numbers (RSSNs) is essential for traceability, quality management, and lifecycle maintenance of railway infrastructure. In practice, embossed serial numbers on concrete surfaces exhibit extremely low discernibility with minimal height variations (<1 mm) and negligible color differentiation. Traditional Red-Green-Blue-based (RGB-based) image enhancement and Optical Character Recognition (OCR) methods face a fundamental limitation: they cannot directly capture the three-dimensional geometric features distinguishing embossed characters from their surroundings. To address this challenge, this study proposes an integrated framework based on line laser height imaging and position prior-guided detection with three key innovations: (1) a cascaded processing framework leverages geometric height information to overcome RGB-based method limitations; (2) a Dual-stage Adaptive Image Enhancement (DAIE) strategy converts 16 binary-digit (bit) height images into optimized 8-bit visualizations by systematically selecting optimal methods: modulo truncation for global structure and Minimum-Maximum (Min-Max) normalization for local detail enhancement; and (3) a Position Prior-guided Spatial Attention (PPSA) Feature Pyramid Network (FPN) integrates statistically-derived position priors to enhance small target detection. Comprehensive validation on 2234 images demonstrates superior performance: 98.2% F1-score and 99.38% recognition accuracy at 27 Frames Per Second (FPS), achieving 2.4% improvement over state-of-the-art methods. Ablation experiments confirm the individual contributions of the PPSA module (4.0%), the Small Target Enhancement (STE) module (1.4%), and the DAIE strategy (3.08%). Field testing in a prefabricated factory validates industrial applicability, providing a scalable technical framework and valuable reference for low-discernibility embossed industrial character recognition. Code is publicly available at <span><span>https://github.com/shipeng38/RSSN-recognition</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 114110"},"PeriodicalIF":8.0,"publicationDate":"2026-02-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175302","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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

Cross-scale hybrid attention network for enhancing performance prediction of modified asphalt binder preparation 改进改性沥青粘结剂制备性能预测的跨尺度混合关注网络

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-02-09 DOI: 10.1016/j.engappai.2026.114106

Jiakang Zhang , Guoan Gan , Kun Long , Allen A. Zhang , Jing Shang , Chuanqi Yan , Changfa Ai

Asphalt materials form the foundation of pavement durability, with styrene–butadiene–styrene (SBS) copolymers widely used to enhance performance. However, the preparation of SBS-modified asphalt (SBSMA) still relies heavily on inefficient trial-and-error approaches. Although artificial intelligence–based methods have been applied to asphalt performance prediction, most existing models directly map preparation parameters to macro-performance, neglecting cross-scale mechanisms linking preparation parameters, micro-properties, and macroscopic behavior. This limitation reduces their robustness and practical applicability in complex material systems. To address this issue, this study proposes a Cross-Scale Hybrid Attention Network (CSA-Net) that explicitly models hierarchical information transfer from preparation parameters to micro-properties and further to macro-performance. CSA-Net adopts a dual-branch architecture: a micro-branch predicts micro-properties using attention-enhanced preparation features, while a macro-branch integrates attention-refined preparation features and predicted micro-features through a second attention module. Joint optimization of micro- and macro-level tasks is achieved via a composite loss function. A comprehensive experimental dataset comprising 864 SBSMA samples was established. Results show that CSA-Net achieves high accuracy in macro-performance prediction, with coefficients of determination (R²) consistently exceeding 0.982, mean absolute percentage errors below 5%, and root mean square errors within experimental uncertainty ranges. Compared with single-scale, multi-scale, and non-attention benchmark models, CSA-Net exhibits improved robustness, as demonstrated by Monte Carlo simulations, with the interquartile range of R² reduced by more than 25%. Shapley additive explanations analysis further reveals meaningful cross-scale relationships between preparation parameters, microstructural evolution, and macroscopic performance. Overall, CSA-Net provides a robust and interpretable framework for intelligent design and performance prediction of modified asphalt binders.

沥青材料是路面耐久性的基础，广泛使用SBS共聚物来提高性能。然而，sbs改性沥青（SBSMA）的制备仍然严重依赖于低效的试错方法。尽管基于人工智能的方法已经应用于沥青性能预测，但大多数现有模型直接将制备参数映射到宏观性能，忽略了连接制备参数、微观性能和宏观行为的跨尺度机制。这一限制降低了它们在复杂材料系统中的稳健性和实用性。为了解决这一问题，本研究提出了一个跨尺度混合注意网络（CSA-Net），该网络明确地模拟了从制备参数到微观特性再到宏观性能的分层信息传递。CSA-Net采用双分支架构：微分支通过注意增强的制备特性预测微观特性，而宏分支通过第二个注意模块集成注意细化的制备特性和预测的微观特性。通过复合损失函数实现微观和宏观任务的联合优化。建立了包含864个SBSMA样本的综合实验数据集。结果表明，CSA-Net在宏观性能预测中具有较高的准确度，决定系数（R2）均大于0.982，平均绝对百分比误差小于5%，均方根误差在实验不确定度范围内。与单尺度、多尺度和非注意力基准模型相比，CSA-Net表现出更好的鲁棒性，蒙特卡罗模拟表明，R2的四分位数范围减小了25%以上。Shapley加性解释分析进一步揭示了制备参数、微观结构演化和宏观性能之间有意义的跨尺度关系。总的来说，CSA-Net为改性沥青粘合剂的智能设计和性能预测提供了一个强大的、可解释的框架。

{"title":"Cross-scale hybrid attention network for enhancing performance prediction of modified asphalt binder preparation","authors":"Jiakang Zhang , Guoan Gan , Kun Long , Allen A. Zhang , Jing Shang , Chuanqi Yan , Changfa Ai","doi":"10.1016/j.engappai.2026.114106","DOIUrl":"10.1016/j.engappai.2026.114106","url":null,"abstract":"<div><div>Asphalt materials form the foundation of pavement durability, with styrene–butadiene–styrene (SBS) copolymers widely used to enhance performance. However, the preparation of SBS-modified asphalt (SBSMA) still relies heavily on inefficient trial-and-error approaches. Although artificial intelligence–based methods have been applied to asphalt performance prediction, most existing models directly map preparation parameters to macro-performance, neglecting cross-scale mechanisms linking preparation parameters, micro-properties, and macroscopic behavior. This limitation reduces their robustness and practical applicability in complex material systems. To address this issue, this study proposes a Cross-Scale Hybrid Attention Network (CSA-Net) that explicitly models hierarchical information transfer from preparation parameters to micro-properties and further to macro-performance. CSA-Net adopts a dual-branch architecture: a micro-branch predicts micro-properties using attention-enhanced preparation features, while a macro-branch integrates attention-refined preparation features and predicted micro-features through a second attention module. Joint optimization of micro- and macro-level tasks is achieved via a composite loss function. A comprehensive experimental dataset comprising 864 SBSMA samples was established. Results show that CSA-Net achieves high accuracy in macro-performance prediction, with coefficients of determination (R<sup>2</sup>) consistently exceeding 0.982, mean absolute percentage errors below 5%, and root mean square errors within experimental uncertainty ranges. Compared with single-scale, multi-scale, and non-attention benchmark models, CSA-Net exhibits improved robustness, as demonstrated by Monte Carlo simulations, with the interquartile range of R<sup>2</sup> reduced by more than 25%. Shapley additive explanations analysis further reveals meaningful cross-scale relationships between preparation parameters, microstructural evolution, and macroscopic performance. Overall, CSA-Net provides a robust and interpretable framework for intelligent design and performance prediction of modified asphalt binders.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 114106"},"PeriodicalIF":8.0,"publicationDate":"2026-02-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174975","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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

Optimizing potential-based reward automata in partially observable reinforcement learning using genetic local search 利用遗传局部搜索优化部分可观察强化学习中基于电位的奖励自动机

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-02-09 DOI: 10.1016/j.engappai.2026.114054

Zhengwei Zhu , Zhixuan Chen , Chenyang Zhu , Wen Si , Fang Wang

Partially observable reinforcement learning extends the reinforcement learning framework to environments in which agents have limited visibility of the state space, making it particularly relevant for applications in robotics and autonomous vehicle navigation. However, a primary challenge in partially observable reinforcement learning is defining effective reward functions that can guide the learning process despite partial observability. To address this challenge, this paper introduces a novel approach for constructing potential-based reward automata by employing genetic local search methods. Specifically, our method constructs these automata from compressed representations of exploration trajectories, which succinctly capture critical decision points and essential state transitions while eliminating redundant steps. By optimizing trajectory samples and shortening agent trajectories to their crucial transitions, our technique significantly reduces computational overhead. Formally, we define the learning objective as an optimization problem aimed at maximizing the log-likelihood of future observations while simultaneously minimizing the structural complexity of the learned reward automata. Furthermore, by incorporating value-based strategies to estimate potential values within the reward automata, our approach improves learning efficiency and facilitates the identification of optimal reward structures. We empirically evaluate our proposed method on seven partially observable grid-world benchmarks. Experimental results demonstrate that our method achieves superior performance relative to state-of-the-art reward automata-based techniques, exhibiting both accelerated learning speeds and higher accumulated rewards. Additionally, our genetic local search algorithm consistently outperforms comparative heuristic methods in terms of learning curves and reward accumulation.

部分可观察强化学习将强化学习框架扩展到代理对状态空间的可见性有限的环境中，使其与机器人和自动车辆导航的应用特别相关。然而，部分可观察强化学习的主要挑战是定义有效的奖励函数，以指导学习过程，尽管部分可观察。为了解决这一问题，本文提出了一种利用遗传局部搜索方法构建基于电位的奖励自动机的新方法。具体来说，我们的方法从勘探轨迹的压缩表示中构建这些自动机，从而简洁地捕获关键决策点和基本状态转换，同时消除冗余步骤。通过优化轨迹样本和缩短智能体轨迹到它们的关键过渡，我们的技术显著降低了计算开销。形式上，我们将学习目标定义为一个优化问题，旨在最大化未来观察的对数似然，同时最小化学习奖励自动机的结构复杂性。此外，通过结合基于价值的策略来估计奖励自动机内的潜在价值，我们的方法提高了学习效率，并有助于识别最佳奖励结构。我们在七个部分可观察的网格世界基准上对我们提出的方法进行了经验评估。实验结果表明，相对于最先进的基于奖励自动机的技术，我们的方法取得了更好的性能，表现出更快的学习速度和更高的累积奖励。此外，我们的遗传局部搜索算法在学习曲线和奖励积累方面始终优于比较启发式方法。

{"title":"Optimizing potential-based reward automata in partially observable reinforcement learning using genetic local search","authors":"Zhengwei Zhu , Zhixuan Chen , Chenyang Zhu , Wen Si , Fang Wang","doi":"10.1016/j.engappai.2026.114054","DOIUrl":"10.1016/j.engappai.2026.114054","url":null,"abstract":"<div><div>Partially observable reinforcement learning extends the reinforcement learning framework to environments in which agents have limited visibility of the state space, making it particularly relevant for applications in robotics and autonomous vehicle navigation. However, a primary challenge in partially observable reinforcement learning is defining effective reward functions that can guide the learning process despite partial observability. To address this challenge, this paper introduces a novel approach for constructing potential-based reward automata by employing genetic local search methods. Specifically, our method constructs these automata from compressed representations of exploration trajectories, which succinctly capture critical decision points and essential state transitions while eliminating redundant steps. By optimizing trajectory samples and shortening agent trajectories to their crucial transitions, our technique significantly reduces computational overhead. Formally, we define the learning objective as an optimization problem aimed at maximizing the log-likelihood of future observations while simultaneously minimizing the structural complexity of the learned reward automata. Furthermore, by incorporating value-based strategies to estimate potential values within the reward automata, our approach improves learning efficiency and facilitates the identification of optimal reward structures. We empirically evaluate our proposed method on seven partially observable grid-world benchmarks. Experimental results demonstrate that our method achieves superior performance relative to state-of-the-art reward automata-based techniques, exhibiting both accelerated learning speeds and higher accumulated rewards. Additionally, our genetic local search algorithm consistently outperforms comparative heuristic methods in terms of learning curves and reward accumulation.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 114054"},"PeriodicalIF":8.0,"publicationDate":"2026-02-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175212","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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

Adaptive multi-agent stock trading decision support system based on deep reinforcement learning 基于深度强化学习的自适应多智能体股票交易决策支持系统

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-02-09 DOI: 10.1016/j.engappai.2026.114130

Xu Yuan , Jiaqiang Wang , Shaokui Gu , Yi Guo , Ange Qi , Shijin Li , Liang Zhao

The stock market is a highly dynamic, complex, and uncertain environment, where traditional investment strategies and technical analysis tools often fail to provide reliable guidance, leading to increased investment risk and uncertainty. This study aims to develop an adaptive multi-agent stock trading decision support system that can effectively respond to volatile market conditions while balancing returns and risk management. We propose a deep reinforcement learning framework based on the Dueling Deep Q-Network (Dueling DQN) algorithm, in which multiple agents independently make optimal trading decisions based on the constructed environment state. The system incorporates a redesigned reward function, a dynamic exploration strategy, and a risk management mechanism to ensure real-time adaptation to market feedback. Extensive experiments on domestic and international market data demonstrate that the proposed system outperforms existing models, effectively responds to market shocks, and exhibits superior adaptability across different market conditions. The proposed multi-agent trading system achieves a robust balance between profitability and risk control, indicating its potential economic value and applicability in dynamic financial markets.

股票市场是一个高度动态、复杂和不确定的环境，传统的投资策略和技术分析工具往往不能提供可靠的指导，导致投资风险和不确定性增加。本研究旨在开发一套自适应的多智能体股票交易决策支持系统，能在平衡收益与风险管理的同时，有效地因应多变的市场环境。我们提出了一种基于Dueling deep Q-Network （Dueling DQN）算法的深度强化学习框架，其中多个智能体根据构建的环境状态独立地做出最优交易决策。该系统整合了重新设计的奖励功能、动态勘探策略和风险管理机制，以确保实时适应市场反馈。国内外市场数据的大量实验表明，该系统优于现有模型，有效应对市场冲击，并在不同市场条件下表现出卓越的适应性。所提出的多智能体交易系统在盈利能力和风险控制之间取得了良好的平衡，显示了其潜在的经济价值和在动态金融市场中的适用性。

{"title":"Adaptive multi-agent stock trading decision support system based on deep reinforcement learning","authors":"Xu Yuan , Jiaqiang Wang , Shaokui Gu , Yi Guo , Ange Qi , Shijin Li , Liang Zhao","doi":"10.1016/j.engappai.2026.114130","DOIUrl":"10.1016/j.engappai.2026.114130","url":null,"abstract":"<div><div>The stock market is a highly dynamic, complex, and uncertain environment, where traditional investment strategies and technical analysis tools often fail to provide reliable guidance, leading to increased investment risk and uncertainty. This study aims to develop an adaptive multi-agent stock trading decision support system that can effectively respond to volatile market conditions while balancing returns and risk management. We propose a deep reinforcement learning framework based on the Dueling Deep Q-Network (Dueling DQN) algorithm, in which multiple agents independently make optimal trading decisions based on the constructed environment state. The system incorporates a redesigned reward function, a dynamic exploration strategy, and a risk management mechanism to ensure real-time adaptation to market feedback. Extensive experiments on domestic and international market data demonstrate that the proposed system outperforms existing models, effectively responds to market shocks, and exhibits superior adaptability across different market conditions. The proposed multi-agent trading system achieves a robust balance between profitability and risk control, indicating its potential economic value and applicability in dynamic financial markets.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 114130"},"PeriodicalIF":8.0,"publicationDate":"2026-02-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175218","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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

Multivariate time series representation learning with multi-task graph neural network 基于多任务图神经网络的多元时间序列表示学习

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-02-09 DOI: 10.1016/j.engappai.2026.113894

Zhihui Gao , Baomin Xu , Jidong Yuan , Jinfeng Wang , Xu Li

Multivariate time series (MTS) representation learning poses a significant challenge in data mining. Current deep learning-based MTS representation methods mostly utilize neural networks to model temporal dependencies within individual univariate sequences, while failing to adequately consider the spatial relationships among different channels within MTS data. While a few methods leverage graph neural networks (GNNs) to model spatial dependencies, but they often do not effectively capture both global and local features simultaneously, potentially limiting the quality of MTS data representations. To overcome these limitations, we present MTGL, a novel Multi-Task Graph Neural Network-based MTS Representation Learning Framework. It leverages MTS reconstruction, global-level graph learning, and local-level graph learning to capture latent spatio-temporal dependencies without relying on predefined graph structures. To obtain global graph-level representations, MTGL performs message-passing and graph pooling operations, and simultaneously leverages a dynamic graph mechanism to capture associations across different windows for local-level representations. By fusing global and local features in a unified framework, MTGL effectively supports a variety of MTS tasks. Extensive experiments show that the proposed method outperforms existing state-of-the-art baselines on benchmark MTS datasets and the tunnel boring machine dataset.

多变量时间序列（MTS）表示学习对数据挖掘提出了重大挑战。当前基于深度学习的MTS表示方法大多利用神经网络来模拟单个单变量序列中的时间依赖性，而未能充分考虑MTS数据中不同通道之间的空间关系。虽然有一些方法利用图神经网络（gnn）来建模空间依赖性，但它们通常不能有效地同时捕获全局和局部特征，这可能会限制MTS数据表示的质量。为了克服这些限制，我们提出了一种新的基于多任务图神经网络的MTS表示学习框架MTGL。它利用MTS重建、全局级图学习和局部级图学习来捕获潜在的时空依赖关系，而不依赖于预定义的图结构。为了获得全局图级表示，MTGL执行消息传递和图池操作，并同时利用动态图机制跨不同窗口捕获局部级表示的关联。通过将全局和局部特征融合在一个统一的框架中，MTGL有效地支持多种MTS任务。大量的实验表明，该方法在基准MTS数据集和隧道掘进机数据集上优于现有的最先进的基线。

{"title":"Multivariate time series representation learning with multi-task graph neural network","authors":"Zhihui Gao , Baomin Xu , Jidong Yuan , Jinfeng Wang , Xu Li","doi":"10.1016/j.engappai.2026.113894","DOIUrl":"10.1016/j.engappai.2026.113894","url":null,"abstract":"<div><div>Multivariate time series (MTS) representation learning poses a significant challenge in data mining. Current deep learning-based MTS representation methods mostly utilize neural networks to model temporal dependencies within individual univariate sequences, while failing to adequately consider the spatial relationships among different channels within MTS data. While a few methods leverage graph neural networks (GNNs) to model spatial dependencies, but they often do not effectively capture both global and local features simultaneously, potentially limiting the quality of MTS data representations. To overcome these limitations, we present <strong>MTGL</strong>, a novel <strong>M</strong>ulti-<strong>T</strong>ask <strong>G</strong>raph Neural Network-based MTS Representation <strong>L</strong>earning Framework. It leverages MTS reconstruction, global-level graph learning, and local-level graph learning to capture latent spatio-temporal dependencies without relying on predefined graph structures. To obtain global graph-level representations, MTGL performs message-passing and graph pooling operations, and simultaneously leverages a dynamic graph mechanism to capture associations across different windows for local-level representations. By fusing global and local features in a unified framework, MTGL effectively supports a variety of MTS tasks. Extensive experiments show that the proposed method outperforms existing state-of-the-art baselines on benchmark MTS datasets and the tunnel boring machine dataset.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 113894"},"PeriodicalIF":8.0,"publicationDate":"2026-02-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175342","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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

Q-learning-driven adaptive rewiring for cooperative control in heterogeneous networks 基于q学习的异构网络协同控制自适应重构

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-02-09 DOI: 10.1016/j.engappai.2026.114024

Yi-Ning Weng , Hsuan-Wei Lee

Cooperation emergence in multi-agent systems represents a fundamental statistical physics problem where microscopic learning rules drive macroscopic collective behavior transitions. We propose a Q-learning-based variant of adaptive rewiring that builds on mechanisms studied in the literature. This method combines temporal difference learning with network restructuring so that agents can optimize strategies and social connections based on interaction histories. Through neighbor-specific Q-learning, agents develop sophisticated partnership management strategies that enable cooperator cluster formation, creating spatial separation between cooperative and defective regions. Using power-law networks that reflect real-world heterogeneous connectivity patterns, we evaluate emergent behaviors under varying rewiring constraint levels, revealing distinct cooperation regimes across parameter space, characterized by qualitative changes in macroscopic cooperation behavior. Our systematic analysis identifies three behavioral regimes: a permissive regime (low constraints) enabling rapid cooperative cluster formation, an intermediate regime with sensitive dependence on dilemma strength, and a patient regime (high constraints) where strategic accumulation gradually optimizes network structure. Comparative analysis against Bush–Mosteller stimulus–response learning demonstrates that Q-learning’s temporal credit assignment capabilities produce superior cooperation outcomes, particularly under intermediate rewiring constraints where long-term relationship assessment becomes crucial. Simulation results show that while moderate constraints create transition-like zones that suppress cooperation, fully adaptive rewiring enhances cooperation levels through systematic exploration of favorable network configurations. Quantitative analysis reveals that increased rewiring frequency drives large-scale cluster formation. Our results establish a new paradigm for understanding intelligence-driven cooperation pattern formation in complex adaptive systems, revealing how machine learning serves as an alternative driving force for spontaneous organization in multi-agent networks.

多智能体系统中的合作出现是一个基本的统计物理问题，微观学习规则驱动宏观集体行为转变。我们提出了一种基于q学习的自适应重新布线变体，该变体建立在文献中研究的机制之上。该方法将时间差异学习与网络重构相结合，使智能体能够基于交互历史优化策略和社会连接。通过特定于邻居的q学习，智能体发展出复杂的伙伴关系管理策略，使合作者集群形成，在合作区域和缺陷区域之间建立空间分离。利用反映现实世界异构连接模式的幂律网络，我们评估了不同重新连接约束水平下的紧急行为，揭示了跨参数空间的不同合作机制，其特征是宏观合作行为的质变。我们的系统分析确定了三种行为机制：允许机制（低约束）能够快速形成合作集群，对困境强度有敏感依赖的中间机制，以及战略积累逐渐优化网络结构的耐心机制（高约束）。与布什-莫斯特勒刺激-反应学习的对比分析表明，q -学习的时间信用分配能力产生了更好的合作结果，特别是在长期关系评估变得至关重要的中间重新布线约束下。仿真结果表明，适度约束会产生抑制合作的过渡区，而完全自适应重布线通过系统地探索有利的网络配置来提高合作水平。定量分析表明，重布线频率的增加推动了大规模星团的形成。我们的研究结果为理解复杂自适应系统中智能驱动的合作模式形成建立了一个新的范式，揭示了机器学习如何作为多智能体网络中自发组织的另一种驱动力。

{"title":"Q-learning-driven adaptive rewiring for cooperative control in heterogeneous networks","authors":"Yi-Ning Weng , Hsuan-Wei Lee","doi":"10.1016/j.engappai.2026.114024","DOIUrl":"10.1016/j.engappai.2026.114024","url":null,"abstract":"<div><div>Cooperation emergence in multi-agent systems represents a fundamental statistical physics problem where microscopic learning rules drive macroscopic collective behavior transitions. We propose a Q-learning-based variant of adaptive rewiring that builds on mechanisms studied in the literature. This method combines temporal difference learning with network restructuring so that agents can optimize strategies and social connections based on interaction histories. Through neighbor-specific Q-learning, agents develop sophisticated partnership management strategies that enable cooperator cluster formation, creating spatial separation between cooperative and defective regions. Using power-law networks that reflect real-world heterogeneous connectivity patterns, we evaluate emergent behaviors under varying rewiring constraint levels, revealing distinct cooperation regimes across parameter space, characterized by qualitative changes in macroscopic cooperation behavior. Our systematic analysis identifies three behavioral regimes: a permissive regime (low constraints) enabling rapid cooperative cluster formation, an intermediate regime with sensitive dependence on dilemma strength, and a patient regime (high constraints) where strategic accumulation gradually optimizes network structure. Comparative analysis against Bush–Mosteller stimulus–response learning demonstrates that Q-learning’s temporal credit assignment capabilities produce superior cooperation outcomes, particularly under intermediate rewiring constraints where long-term relationship assessment becomes crucial. Simulation results show that while moderate constraints create transition-like zones that suppress cooperation, fully adaptive rewiring enhances cooperation levels through systematic exploration of favorable network configurations. Quantitative analysis reveals that increased rewiring frequency drives large-scale cluster formation. Our results establish a new paradigm for understanding intelligence-driven cooperation pattern formation in complex adaptive systems, revealing how machine learning serves as an alternative driving force for spontaneous organization in multi-agent networks.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 114024"},"PeriodicalIF":8.0,"publicationDate":"2026-02-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174978","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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

Semi-supervised vessel trajectory analysis for unregulated fishing activity detection 无管制捕捞活动检测的半监督船舶轨迹分析

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-02-09 DOI: 10.1016/j.engappai.2026.113933

Agam Sanghera , Paramveer Singh , Elaine Chu , Sumin Leem , Ruizhi Li , Sogol Ghattan , Andy Man Yeung Tai

Background/Problem

Maritime crimes such as Illegal, Unreported, and Unregulated (IUU) fishing, piracy, and smuggling pose significant threats to marine ecosystems, trade, and coastal security, especially in developing regions. Automatic Identification System (AIS) data offers a scalable solution for vessel monitoring, but the use of fully supervised machine learning models is constrained by the substantial manual effort and expert input required to label training data.

Methods

To address this challenge, authors propose a semi-supervised machine learning pipeline that classifies vessel activities from AIS data without relying on pre-labeled datasets. Our approach leverages scaled geospatial and temporal features, including latitude, longitude, speed, and time difference, to train multiple Hidden Markov Models (HMMs) on trajectory segments. These segments are then grouped using similarity-based K-means clustering and subsequently classified with supervised models, including Random Forest and Long Short-Term Memory (LSTM) networks. The pipeline effectively identifies and labels maritime activities such as sailing, fishing, idling, and other activities.

Results/conclusions

Experiments were conducted on a dataset comprising 156,379 AIS points, partitioned into training and test sets. The LSTM-based supervised model achieved an F1 score of 0.86 on the local test set, while the end-to-end pipeline achieved an F1 score of 0.5 on a global evaluation set. These results demonstrate the feasibility of automating maritime activity classification through artificial intelligence and hybrid learning, offering a scalable solution for real-world maritime surveillance.

背景/问题非法、不报告和不管制（IUU）捕鱼、海盗和走私等海上犯罪对海洋生态系统、贸易和沿海安全构成重大威胁，特别是在发展中地区。自动识别系统（AIS）数据为船舶监测提供了一种可扩展的解决方案，但是完全监督机器学习模型的使用受到标记训练数据所需的大量人工和专家输入的限制。为了解决这一挑战，作者提出了一种半监督机器学习管道，可以从AIS数据中对船舶活动进行分类，而不依赖于预先标记的数据集。我们的方法利用缩放的地理空间和时间特征，包括纬度、经度、速度和时差，在轨迹段上训练多个隐马尔可夫模型（hmm）。然后使用基于相似性的k均值聚类对这些片段进行分组，随后使用监督模型进行分类，包括随机森林和长短期记忆（LSTM）网络。该管道可以有效地识别和标记海上活动，如航行、捕鱼、空转等活动。结果/结论在包含156,379个AIS点的数据集上进行了实验，分为训练集和测试集。基于lstm的监督模型在局部测试集上的F1得分为0.86，而端到端管道在全局评估集上的F1得分为0.5。这些结果证明了通过人工智能和混合学习实现海上活动自动化分类的可行性，为现实世界的海上监视提供了可扩展的解决方案。

{"title":"Semi-supervised vessel trajectory analysis for unregulated fishing activity detection","authors":"Agam Sanghera , Paramveer Singh , Elaine Chu , Sumin Leem , Ruizhi Li , Sogol Ghattan , Andy Man Yeung Tai","doi":"10.1016/j.engappai.2026.113933","DOIUrl":"10.1016/j.engappai.2026.113933","url":null,"abstract":"<div><h3>Background/Problem</h3><div>Maritime crimes such as Illegal, Unreported, and Unregulated (IUU) fishing, piracy, and smuggling pose significant threats to marine ecosystems, trade, and coastal security, especially in developing regions. Automatic Identification System (AIS) data offers a scalable solution for vessel monitoring, but the use of fully supervised machine learning models is constrained by the substantial manual effort and expert input required to label training data.</div></div><div><h3>Methods</h3><div>To address this challenge, authors propose a semi-supervised machine learning pipeline that classifies vessel activities from AIS data without relying on pre-labeled datasets. Our approach leverages scaled geospatial and temporal features, including latitude, longitude, speed, and time difference, to train multiple Hidden Markov Models (HMMs) on trajectory segments. These segments are then grouped using similarity-based K-means clustering and subsequently classified with supervised models, including Random Forest and Long Short-Term Memory (LSTM) networks. The pipeline effectively identifies and labels maritime activities such as sailing, fishing, idling, and other activities.</div></div><div><h3>Results/conclusions</h3><div>Experiments were conducted on a dataset comprising 156,379 AIS points, partitioned into training and test sets. The LSTM-based supervised model achieved an F1 score of 0.86 on the local test set, while the end-to-end pipeline achieved an F1 score of 0.5 on a global evaluation set. These results demonstrate the feasibility of automating maritime activity classification through artificial intelligence and hybrid learning, offering a scalable solution for real-world maritime surveillance.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 113933"},"PeriodicalIF":8.0,"publicationDate":"2026-02-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175351","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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

Dual-attention semi-cycled generative adversarial network data augmentation structure for gearbox fault diagnosis using infrared thermal images 基于红外热图像的齿轮箱故障诊断双注意力半循环生成对抗网络数据增强结构

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-02-09 DOI: 10.1016/j.engappai.2026.114118

Zhenlong Chen , Xiao Zhuang , Di Zhou , Weifang Sun , Jing Cai , Jiawei Xiang

In the real world, the data imbalance problem is an everlasting challenge in the field of knowledge-based fault diagnosis. The cycle-consistent generative adversarial network (CycleGAN) has been widely used in sample generation tasks with good results, but still faces the discrepancy between generator-generated fault samples and synthesized fault samples, which leads to the joint participation of the same degenerate branch affecting the final performance, resulting in the tendency to produce artifacts in difficult scenarios. To better solve the data imbalanced problem in infrared image-based fault diagnosis, in this paper, a novel Dual-attention Semi-Cycled Data Augmentation Structure using generative adversarial network (DASCGAN) is proposed to generate high-quality infrared thermal images to solve the fault diagnosis of gearbox under data imbalance. Firstly, the Dual-attention Semi-Cycled Data Augmentation Structure is constructed, which includes the forward and backward semi-cycle sub-network to generate diverse infrared images. The two semi-cycle sub-networks consist of two independent fault-free generation branches and a shared fault generation branch. Secondly, Convolutional Block Attention Module (CBAM) is embedded in the fault-free generation branch to increase the attention of the fault part by suppressing irrelevant features. Thirdly, self-attention module is embedded in the fault generation branch to increase the attention of the fault generation branch on the global hot pixels. Finally, comparison experiments are conducted. Experimental results show that the proposed DASCGAN method outperforms other benchmark generative adversarial networks. The proposed DASCGAN can effectively solve the data imbalance so as to improve the reliability and accuracy of gearbox fault diagnosis.

在现实世界中，数据不平衡问题一直是基于知识的故障诊断领域面临的难题。循环一致生成对抗网络（CycleGAN）在样本生成任务中得到了广泛的应用，并取得了良好的效果，但仍然面临着生成器生成的故障样本与合成的故障样本之间的差异，导致同一退化分支的共同参与影响最终性能，在困难场景中容易产生伪影。为了更好地解决基于红外图像的故障诊断中的数据不平衡问题，本文提出了一种新的基于生成对抗网络的双注意力半循环数据增强结构（DASCGAN）来生成高质量的红外热图像，以解决数据不平衡下的齿轮箱故障诊断问题。首先，构建了双关注半周期数据增强结构，该结构包括正向半周期子网络和反向半周期子网络，用于生成多种红外图像；两个半周期子网由两个独立的无故障发电支路和一个共享的故障发电支路组成。其次，在无故障生成分支中嵌入卷积块注意模块（CBAM），通过抑制不相关特征来增加对故障部分的关注；第三，在故障生成分支中嵌入自关注模块，增加故障生成分支对全局热像素的关注；最后进行了对比实验。实验结果表明，所提出的DASCGAN方法优于其他基准生成对抗网络。所提出的DASCGAN可以有效地解决数据不平衡问题，从而提高齿轮箱故障诊断的可靠性和准确性。

{"title":"Dual-attention semi-cycled generative adversarial network data augmentation structure for gearbox fault diagnosis using infrared thermal images","authors":"Zhenlong Chen , Xiao Zhuang , Di Zhou , Weifang Sun , Jing Cai , Jiawei Xiang","doi":"10.1016/j.engappai.2026.114118","DOIUrl":"10.1016/j.engappai.2026.114118","url":null,"abstract":"<div><div>In the real world, the data imbalance problem is an everlasting challenge in the field of knowledge-based fault diagnosis. The cycle-consistent generative adversarial network (CycleGAN) has been widely used in sample generation tasks with good results, but still faces the discrepancy between generator-generated fault samples and synthesized fault samples, which leads to the joint participation of the same degenerate branch affecting the final performance, resulting in the tendency to produce artifacts in difficult scenarios. To better solve the data imbalanced problem in infrared image-based fault diagnosis, in this paper, a novel Dual-attention Semi-Cycled Data Augmentation Structure using generative adversarial network (DASCGAN) is proposed to generate high-quality infrared thermal images to solve the fault diagnosis of gearbox under data imbalance. Firstly, the Dual-attention Semi-Cycled Data Augmentation Structure is constructed, which includes the forward and backward semi-cycle sub-network to generate diverse infrared images. The two semi-cycle sub-networks consist of two independent fault-free generation branches and a shared fault generation branch. Secondly, Convolutional Block Attention Module (CBAM) is embedded in the fault-free generation branch to increase the attention of the fault part by suppressing irrelevant features. Thirdly, self-attention module is embedded in the fault generation branch to increase the attention of the fault generation branch on the global hot pixels. Finally, comparison experiments are conducted. Experimental results show that the proposed DASCGAN method outperforms other benchmark generative adversarial networks. The proposed DASCGAN can effectively solve the data imbalance so as to improve the reliability and accuracy of gearbox fault diagnosis.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 114118"},"PeriodicalIF":8.0,"publicationDate":"2026-02-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175211","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 learning-based coke dry quenching material location prediction using physical information reconstruction features 基于物理信息重构特征的深度学习焦炭干淬物料位置预测

IF 8 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Engineering Applications of Artificial Intelligence

Pub Date : 2026-02-09 DOI: 10.1016/j.engappai.2026.114117

Xinyang Meng , Keliang Pang , Zhiyuan Gu , Youzhi Zheng , Fujun Liu , Chaoran Wan , Haotian Wu , Minmin Sun , Hua Zhao

Coke dry quenching (CDQ) is a common, environmentally friendly technology applied in iron and steel production and plays an important role in improving coke quality as well as in emission reduction and pollution reduction. Material location prediction is crucial for ensuring the stable operation of dry quenching systems. In this paper, we propose a novel artificial intelligence approach for predicting the location of coke materials in CDQ furnaces by incorporating a method known as physical information feature reconstruction (PIFR). This method integrates physical a priori knowledge (such as the law of mass conservation and furnace structural characteristics) into the feature engineering process, effectively improving the accuracy and stability of time-series predictions in both single-step and multistep forecasting tasks. The experimental results demonstrate that PIFR significantly enhances the performance of various deep learning models. Specifically, for the long short-term memory model, the mean squared error and mean absolute error decreased by 51.25% and 37.63%, respectively, whereas the coefficient of determination increased to 0.941. Moreover, PIFR effectively mitigates issues commonly encountered in multi-step prediction, such as cumulative error and prediction curve flattening. The application of PIFR not only improves the accuracy of the model but also significantly enhances its generalization capability.

焦炭干熄法是钢铁生产中常用的一种环保技术，在提高焦炭质量、减少排放和污染方面具有重要作用。物料位置预测是保证干淬火系统稳定运行的关键。在本文中，我们提出了一种新的人工智能方法，通过结合一种称为物理信息特征重构（PIFR）的方法来预测CDQ炉中焦炭材料的位置。该方法将物理先验知识（如质量守恒定律、炉膛结构特征等）整合到特征工程过程中，有效提高了单步和多步预测任务中时间序列预测的准确性和稳定性。实验结果表明，PIFR显著提高了各种深度学习模型的性能。其中，长短期记忆模型的均方误差和平均绝对误差分别下降了51.25%和37.63%，而决定系数增加到0.941。此外，PIFR有效地缓解了多步预测中常见的问题，如累积误差和预测曲线平坦化。PIFR的应用不仅提高了模型的精度，而且显著增强了模型的泛化能力。

{"title":"Deep learning-based coke dry quenching material location prediction using physical information reconstruction features","authors":"Xinyang Meng , Keliang Pang , Zhiyuan Gu , Youzhi Zheng , Fujun Liu , Chaoran Wan , Haotian Wu , Minmin Sun , Hua Zhao","doi":"10.1016/j.engappai.2026.114117","DOIUrl":"10.1016/j.engappai.2026.114117","url":null,"abstract":"<div><div>Coke dry quenching (CDQ) is a common, environmentally friendly technology applied in iron and steel production and plays an important role in improving coke quality as well as in emission reduction and pollution reduction. Material location prediction is crucial for ensuring the stable operation of dry quenching systems. In this paper, we propose a novel artificial intelligence approach for predicting the location of coke materials in CDQ furnaces by incorporating a method known as physical information feature reconstruction (PIFR). This method integrates physical <em>a priori</em> knowledge (such as the law of mass conservation and furnace structural characteristics) into the feature engineering process, effectively improving the accuracy and stability of time-series predictions in both single-step and multistep forecasting tasks. The experimental results demonstrate that PIFR significantly enhances the performance of various deep learning models. Specifically, for the long short-term memory model, the mean squared error and mean absolute error decreased by 51.25% and 37.63%, respectively, whereas the coefficient of determination increased to 0.941. Moreover, PIFR effectively mitigates issues commonly encountered in multi-step prediction, such as cumulative error and prediction curve flattening. The application of PIFR not only improves the accuracy of the model but also significantly enhances its generalization capability.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"169 ","pages":"Article 114117"},"PeriodicalIF":8.0,"publicationDate":"2026-02-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175344","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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