Computers & Chemical Engineering最新文献_第6页

MINLP-based integrated modeling and multi-period optimization of mass-energy coupled FCC-steam systems with carbon-cost-oriented economic objective 以碳成本为经济目标的基于minlp的质能耦合FCC-steam系统集成建模与多周期优化

IF 3.9 2区工程技术 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Chemical Engineering

Pub Date : 2025-11-24 DOI: 10.1016/j.compchemeng.2025.109503

Jian Long , Bishi Zhao , Kai Deng , Cheng Huang , Chen Fan

With increasing global pressure to decarbonize the energy and chemical industries, the oil refining sector is undergoing a critical transformation toward green and low-carbon development. As one of the core oil refining units, the fluid catalytic cracking (FCC) process is complex. Meanwhile, it has high energy consumption and large carbon emissions. Separate optimization leads to the loss of energy and quality synergy. To address the issue of simultaneous energy and quality losses resulting from the separate optimization of the FCC and steam systems, this study models and optimizes the multi-cycle energy and quality coupling of catalytic cracking process and steam system collaboration. Based on the deep coupling of the cracking reaction and the dynamic transmission characteristics of the steam pipeline network, a multi-time-scale coupling model is established to reveal the interaction mechanism between the device and the steam system. This work develops a mathematical framework based on mixed-integer linear optimization, which aims to enhance the overall economic performance of the integrated plant, integrating the topological constraints of the pipeline network, the variable operating conditions characteristics of the equipment, and the discrete start-stop logic. Through case verification and system decoupling comparative experiments, the revenue increase of the global optimization scheme with energy and quality coupling reached 41.2 %, proving that the proposed method can effectively improve energy efficiency in the optimization scheme under the actual refinery.

随着全球能源和化工行业脱碳压力的加大，炼油行业正在经历一场向绿色低碳发展的关键转型。流化催化裂化（FCC）作为核心炼油装置之一，工艺复杂。同时，它的能耗高，碳排放量大。单独优化导致能量和质量协同的损失。针对催化裂化过程中催化裂化过程与蒸汽系统分别优化导致的能量和质量同时损失的问题，本研究对催化裂化过程与蒸汽系统协同的多循环能量和质量耦合进行了建模和优化。基于裂化反应的深度耦合和蒸汽管网的动态传输特性，建立了多时间尺度耦合模型，揭示了装置与蒸汽系统的相互作用机理。本工作开发了一个基于混合整数线性优化的数学框架，旨在通过集成管网的拓扑约束、设备的可变运行条件特征和离散启停逻辑，提高综合工厂的整体经济性能。通过实例验证和系统解耦对比实验，能量与质量耦合的全局优化方案的收益增幅达到41.2%，证明所提方法能有效提高实际炼油厂优化方案的能效。

{"title":"MINLP-based integrated modeling and multi-period optimization of mass-energy coupled FCC-steam systems with carbon-cost-oriented economic objective","authors":"Jian Long , Bishi Zhao , Kai Deng , Cheng Huang , Chen Fan","doi":"10.1016/j.compchemeng.2025.109503","DOIUrl":"10.1016/j.compchemeng.2025.109503","url":null,"abstract":"<div><div>With increasing global pressure to decarbonize the energy and chemical industries, the oil refining sector is undergoing a critical transformation toward green and low-carbon development. As one of the core oil refining units, the fluid catalytic cracking (FCC) process is complex. Meanwhile, it has high energy consumption and large carbon emissions. Separate optimization leads to the loss of energy and quality synergy. To address the issue of simultaneous energy and quality losses resulting from the separate optimization of the FCC and steam systems, this study models and optimizes the multi-cycle energy and quality coupling of catalytic cracking process and steam system collaboration. Based on the deep coupling of the cracking reaction and the dynamic transmission characteristics of the steam pipeline network, a multi-time-scale coupling model is established to reveal the interaction mechanism between the device and the steam system. This work develops a mathematical framework based on mixed-integer linear optimization, which aims to enhance the overall economic performance of the integrated plant, integrating the topological constraints of the pipeline network, the variable operating conditions characteristics of the equipment, and the discrete start-stop logic. Through case verification and system decoupling comparative experiments, the revenue increase of the global optimization scheme with energy and quality coupling reached 41.2 %, proving that the proposed method can effectively improve energy efficiency in the optimization scheme under the actual refinery.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"206 ","pages":"Article 109503"},"PeriodicalIF":3.9,"publicationDate":"2025-11-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145691684","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Oscillation analysis using Retrieval-Augmented Generation 基于检索增广生成的振荡分析

IF 3.9 2区工程技术 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Chemical Engineering

Pub Date : 2025-11-22 DOI: 10.1016/j.compchemeng.2025.109489

Abhijeet Singh , Mohammadhossein Modir Rousta , Biao Huang

Industrial control system oscillations pose significant operational challenges, causing economic losses through energy waste, equipment degradation, and reduced product quality. Traditional detection methods rely heavily on manual expert analysis, creating scalability constraints in facilities with thousands of control loops. This paper presents a novel framework integrating Large Language Models (LLMs) with specialized oscillation detection toolboxes through Retrieval-Augmented Generation (RAG). The system features a command-line interface enabling seamless programmatic interaction between LLMs and analytical tools, eliminating GUI dependencies. A domain-specific RAG architecture combines real-time analytical outputs with technical knowledge repositories, while natural language processing capabilities allow industrial personnel to query systems using everyday language. The framework incorporates triangle-like shape detection algorithms enhanced by intelligent LLM interpretation. Validation using industrial datasets from refinery operations, the International Stiction Database, and the Tennessee Eastman Process benchmark demonstrates substantial performance improvements, achieving excellent classification accuracy and correlation values exceeding 0.96, effectively democratizing access to advanced oscillation analysis capabilities.

工业控制系统振荡带来了重大的操作挑战，通过能源浪费、设备退化和产品质量降低造成经济损失。传统的检测方法严重依赖于人工专家分析，在具有数千个控制回路的设施中产生可扩展性限制。本文提出了一种通过检索增强生成（RAG）将大型语言模型（llm）与专门的振荡检测工具箱集成在一起的新框架。该系统具有命令行界面，可实现llm和分析工具之间的无缝编程交互，消除了对GUI的依赖。特定于领域的RAG体系结构将实时分析输出与技术知识库结合在一起，而自然语言处理功能允许工业人员使用日常语言查询系统。该框架结合了由智能LLM解释增强的类三角形形状检测算法。使用来自炼油厂作业的工业数据集、国际粘度数据库和田纳西伊士曼过程基准进行验证，证明了显著的性能改进，实现了出色的分类精度和超过0.96的相关值，有效地实现了高级振荡分析功能的大众化。

{"title":"Oscillation analysis using Retrieval-Augmented Generation","authors":"Abhijeet Singh , Mohammadhossein Modir Rousta , Biao Huang","doi":"10.1016/j.compchemeng.2025.109489","DOIUrl":"10.1016/j.compchemeng.2025.109489","url":null,"abstract":"<div><div>Industrial control system oscillations pose significant operational challenges, causing economic losses through energy waste, equipment degradation, and reduced product quality. Traditional detection methods rely heavily on manual expert analysis, creating scalability constraints in facilities with thousands of control loops. This paper presents a novel framework integrating Large Language Models (LLMs) with specialized oscillation detection toolboxes through Retrieval-Augmented Generation (RAG). The system features a command-line interface enabling seamless programmatic interaction between LLMs and analytical tools, eliminating GUI dependencies. A domain-specific RAG architecture combines real-time analytical outputs with technical knowledge repositories, while natural language processing capabilities allow industrial personnel to query systems using everyday language. The framework incorporates triangle-like shape detection algorithms enhanced by intelligent LLM interpretation. Validation using industrial datasets from refinery operations, the International Stiction Database, and the Tennessee Eastman Process benchmark demonstrates substantial performance improvements, achieving excellent classification accuracy and correlation values exceeding 0.96, effectively democratizing access to advanced oscillation analysis capabilities.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"206 ","pages":"Article 109489"},"PeriodicalIF":3.9,"publicationDate":"2025-11-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145594796","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

The incorporation of qualitative knowledge in hybrid modeling 混合建模中定性知识的结合

IF 3.9 2区工程技术 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Chemical Engineering

Pub Date : 2025-11-18 DOI: 10.1016/j.compchemeng.2025.109484

Elia Arnese-Feffin , Nidhish Sagar , Luis A. Briceno-Mena , Birgit Braun , Ivan Castillo , Caterina Rizzo , Linh Bui , Jinsuo Xu , Leo H. Chiang , Richard D. Braatz

Data-driven modeling enables the pursuit of model-based solutions when no or partial knowledge is available on the process of interest. The typically poor generalization performance of data-driven models can be improved by incorporating all available knowledge into the model development workflow, i.e. by hybrid modeling. However, how to incorporate this knowledge and what kind of knowledge to use have been strictly problem-dependent questions: whether a general framework for hybrid modeling can be devised remains an open research topic. In this article, we make the first step towards such an objective: we propose a method relying on theoretically sound modification of the objective function of data-driven models by incorporation of constraints derived from process knowledge, an approach that can be readily implemented in existing software packages with minimal overhead. We focus on an underutilized source of knowledge, i.e. qualitative knowledge. The proposed approach is demonstrated in two case studies, where we employ qualitative and quantitative process knowledge with varying degrees of complexity and discuss their effectiveness. Results show promising performance, paving the way for a truly general hybrid modeling framework.

当感兴趣的过程没有或只有部分知识可用时，数据驱动的建模支持追求基于模型的解决方案。通过将所有可用的知识合并到模型开发工作流中，即通过混合建模，可以改善数据驱动模型的典型的较差的泛化性能。然而，如何整合这些知识以及使用什么样的知识一直是严格依赖于问题的问题：是否可以设计出混合建模的通用框架仍然是一个开放的研究课题。在本文中，我们向这样一个目标迈出了第一步：我们提出了一种方法，该方法依赖于数据驱动模型的目标函数的理论上合理的修改，通过合并来自过程知识的约束，这种方法可以很容易地在现有软件包中以最小的开销实现。我们专注于未充分利用的知识来源，即定性知识。所提出的方法在两个案例研究中得到了证明，在这些案例研究中，我们采用了不同复杂程度的定性和定量过程知识，并讨论了它们的有效性。结果显示了良好的性能，为真正通用的混合建模框架铺平了道路。

{"title":"The incorporation of qualitative knowledge in hybrid modeling","authors":"Elia Arnese-Feffin , Nidhish Sagar , Luis A. Briceno-Mena , Birgit Braun , Ivan Castillo , Caterina Rizzo , Linh Bui , Jinsuo Xu , Leo H. Chiang , Richard D. Braatz","doi":"10.1016/j.compchemeng.2025.109484","DOIUrl":"10.1016/j.compchemeng.2025.109484","url":null,"abstract":"<div><div>Data-driven modeling enables the pursuit of model-based solutions when no or partial knowledge is available on the process of interest. The typically poor generalization performance of data-driven models can be improved by incorporating all available knowledge into the model development workflow, i.e. by hybrid modeling. However, how to incorporate this knowledge and what kind of knowledge to use have been strictly problem-dependent questions: whether a general framework for hybrid modeling can be devised remains an open research topic. In this article, we make the first step towards such an objective: we propose a method relying on theoretically sound modification of the objective function of data-driven models by incorporation of constraints derived from process knowledge, an approach that can be readily implemented in existing software packages with minimal overhead. We focus on an underutilized source of knowledge, i.e. <em>qualitative</em> knowledge. The proposed approach is demonstrated in two case studies, where we employ qualitative and quantitative process knowledge with varying degrees of complexity and discuss their effectiveness. Results show promising performance, paving the way for a truly general hybrid modeling framework.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"205 ","pages":"Article 109484"},"PeriodicalIF":3.9,"publicationDate":"2025-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145620087","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Multi-scale residual attention networks with spatio-temporal attention co-optimization for industrial fault diagnosis 基于时空关注协同优化的多尺度剩余关注网络工业故障诊断

IF 3.9 2区工程技术 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Chemical Engineering

Pub Date : 2025-11-17 DOI: 10.1016/j.compchemeng.2025.109482

Youqiang Chen , Ridong Zhang , Furong Gao

Traditional fault diagnosis has limitations in dynamic condition adaptation and multimodal feature decoupling, while existing deep learning methods are limited by the localization of time series modeling, noise sensitivity and insufficient efficiency of multiscale feature fusion. Aiming at the above challenges, this paper proposes a multi-scale residual attention fault diagnosis network (MRAFDN) through hierarchical dynamic feature enhancement and spatio-temporal attention co-optimization mechanism. The innovation of this study lies in constructing a dual fusion framework integrating Dynamic Multi-Scale Channel Attention (MSCA) and Dilated Spatial Attention (DSA). Employing a 1D Residual Network (1D-ResNet) as the backbone architecture, residual connections preserve low-level information from raw signals, while cascaded optimization with dual attention mechanisms forms a comprehensive lifecycle fault characterization system spanning from micro fluctuations to macro drifts. This methodology theoretically addresses fundamental challenges including noise-feature coupling, coexistence of multiple fault modes and adaptability to dynamic operational conditions. The proposed method demonstrates significant robustness under strong noise for multi-class fault modes in TE chemical processes and industrial coke furnaces.

传统的故障诊断在动态条件自适应和多模态特征解耦方面存在局限性，而现有的深度学习方法则受到时间序列建模局部化、噪声敏感性和多尺度特征融合效率不足的限制。针对上述挑战，本文提出了一种基于分层动态特征增强和时空注意力协同优化机制的多尺度剩余注意力故障诊断网络（MRAFDN）。本研究的创新点在于构建了动态多尺度通道注意（MSCA）和扩展空间注意（DSA）的双融合框架。残差连接采用一维残差网络（1D- resnet）作为主干架构，保留了原始信号中的低级信息，而采用双注意机制的级联优化形成了从微观波动到宏观漂移的综合生命周期故障表征系统。该方法从理论上解决了包括噪声-特征耦合、多种故障模式共存以及对动态运行条件的适应性在内的基本挑战。该方法对TE化工过程和工业焦炉的多类故障模式在强噪声下具有较强的鲁棒性。

{"title":"Multi-scale residual attention networks with spatio-temporal attention co-optimization for industrial fault diagnosis","authors":"Youqiang Chen , Ridong Zhang , Furong Gao","doi":"10.1016/j.compchemeng.2025.109482","DOIUrl":"10.1016/j.compchemeng.2025.109482","url":null,"abstract":"<div><div>Traditional fault diagnosis has limitations in dynamic condition adaptation and multimodal feature decoupling, while existing deep learning methods are limited by the localization of time series modeling, noise sensitivity and insufficient efficiency of multiscale feature fusion. Aiming at the above challenges, this paper proposes a multi-scale residual attention fault diagnosis network (MRAFDN) through hierarchical dynamic feature enhancement and spatio-temporal attention co-optimization mechanism. The innovation of this study lies in constructing a dual fusion framework integrating Dynamic Multi-Scale Channel Attention (MSCA) and Dilated Spatial Attention (DSA). Employing a 1D Residual Network (1D-ResNet) as the backbone architecture, residual connections preserve low-level information from raw signals, while cascaded optimization with dual attention mechanisms forms a comprehensive lifecycle fault characterization system spanning from micro fluctuations to macro drifts. This methodology theoretically addresses fundamental challenges including noise-feature coupling, coexistence of multiple fault modes and adaptability to dynamic operational conditions. The proposed method demonstrates significant robustness under strong noise for multi-class fault modes in TE chemical processes and industrial coke furnaces.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"205 ","pages":"Article 109482"},"PeriodicalIF":3.9,"publicationDate":"2025-11-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145620086","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Multi-criteria decision-making and multi-objective optimization of a sustainable bio-based isopropanol supply chain 可持续生物基异丙醇供应链的多准则决策和多目标优化

IF 3.9 2区工程技术 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Chemical Engineering

Pub Date : 2025-11-17 DOI: 10.1016/j.compchemeng.2025.109487

Ching-Mei Wen, Marianthi Ierapetritou

This study focuses the multi-objective optimization and multi-criteria decision-making (MCDM) framework for designing a sustainable bio-based isopropanol (bio-IPA) supply chain under uncertainty, using sugar beet as the primary feedstock. By applying a two-stage stochastic programming model, the research optimizes the five-tier supply chain from sugar beet farms through storage, sugar plants, IPA production, and retail distribution across Minnesota. The study addresses economic, environmental, and supply chain performance objectives while accounting for uncertainties in biomass yield and market demand fluctuations. Results shows a cost-optimal supply chain configuration that can adapt to demand surges, achieving up to 94 % service levels when a $2.5/kg penalty for unsatisfied demand is applied. The analysis highlights the dominant influence of biomass cost, which accounts for approximately 65 % of total production costs, underscoring its critical role in supply chain economics. As environmental constraints tighten (e.g., greenhouse gas emission caps), the system experiences rising unsatisfied demand penalties and operational challenges. Furthermore, the study applies MCDM techniques, including the Weighted Product Method and data-driven weighting methods such as the coefficient of variation and interval entropy, to rank alternative configurations. Ternary plots, reflect the economic, environmental, and operational (performance-based) dimensions, are proposed as a tool to visualize trade-offs across economic, environmental, and operational dimensions, enhancing informed decision-making by exploring a diverse range of weight distributions that reflect varying stakeholder values.

本研究以甜菜为主要原料，采用多目标优化和多准则决策（MCDM）框架，设计了不确定条件下可持续的生物基异丙醇（bio-IPA）供应链。通过应用两阶段随机规划模型，该研究优化了从甜菜农场到储存、糖厂、IPA生产和明尼苏达州零售分销的五层供应链。该研究解决了经济、环境和供应链绩效目标，同时考虑了生物质产量和市场需求波动的不确定性。结果表明，成本最优的供应链配置可以适应需求激增，当对未满足的需求施加2.5美元/公斤的罚款时，服务水平可达94%。该分析强调了生物质成本的主要影响，约占总生产成本的65%，强调了其在供应链经济中的关键作用。随着环境约束的收紧（例如，温室气体排放上限），系统经历了不断增加的未满足需求的惩罚和操作挑战。此外，该研究应用MCDM技术，包括加权乘积法和数据驱动的加权方法，如变异系数和区间熵，对可选配置进行排序。三元图反映了经济、环境和运营（基于绩效的）维度，被提议作为一种工具，将经济、环境和运营维度之间的权衡可视化，通过探索反映不同利益相关者价值的各种权重分布来增强明智的决策。

{"title":"Multi-criteria decision-making and multi-objective optimization of a sustainable bio-based isopropanol supply chain","authors":"Ching-Mei Wen, Marianthi Ierapetritou","doi":"10.1016/j.compchemeng.2025.109487","DOIUrl":"10.1016/j.compchemeng.2025.109487","url":null,"abstract":"<div><div>This study focuses the multi-objective optimization and multi-criteria decision-making (MCDM) framework for designing a sustainable bio-based isopropanol (bio-IPA) supply chain under uncertainty, using sugar beet as the primary feedstock. By applying a two-stage stochastic programming model, the research optimizes the five-tier supply chain from sugar beet farms through storage, sugar plants, IPA production, and retail distribution across Minnesota. The study addresses economic, environmental, and supply chain performance objectives while accounting for uncertainties in biomass yield and market demand fluctuations. Results shows a cost-optimal supply chain configuration that can adapt to demand surges, achieving up to 94 % service levels when a $2.5/kg penalty for unsatisfied demand is applied. The analysis highlights the dominant influence of biomass cost, which accounts for approximately 65 % of total production costs, underscoring its critical role in supply chain economics. As environmental constraints tighten (e.g., greenhouse gas emission caps), the system experiences rising unsatisfied demand penalties and operational challenges. Furthermore, the study applies MCDM techniques, including the Weighted Product Method and data-driven weighting methods such as the coefficient of variation and interval entropy, to rank alternative configurations. Ternary plots, reflect the economic, environmental, and operational (performance-based) dimensions, are proposed as a tool to visualize trade-offs across economic, environmental, and operational dimensions, enhancing informed decision-making by exploring a diverse range of weight distributions that reflect varying stakeholder values.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"205 ","pages":"Article 109487"},"PeriodicalIF":3.9,"publicationDate":"2025-11-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145576349","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Oil production forecasting using temporal Kolmogorov–Arnold networks 利用时间Kolmogorov-Arnold网络进行石油产量预测

IF 3.9 2区工程技术 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Chemical Engineering

Pub Date : 2025-11-14 DOI: 10.1016/j.compchemeng.2025.109483

Mengze Zheng, Tao Zhang, Jing Cao, Zhong Chen, Jian Zou

Given the challenges associated with resource depletion and market volatility, accurate oil production forecasting has become a critical component for optimizing oilfield development and enhancing decision-making processes. In recent years, various machine learning and deep learning methods have been widely adopted. However, these approaches still exhibit significant limitations in terms of accuracy and generalizability, often failing to fully capture the complexities of dynamic reservoir environments and multivariate datasets. To address these challenges, we propose the temporal Kolmogorov–Arnold networks (TKAN), a novel deep learning architecture specifically designed for multivariate time-series forecasting in the context of oil production. TKAN integrates Kolmogorov–Arnold decomposition with adaptive spline-enhanced activation functions, enabling the model to effectively capture nonlinear relationships and temporal dependencies. This leads to substantial improvements over conventional techniques when dealing with noisy and dynamic datasets. In the experimental section, the proposed TKAN model is utilized to predict oil production in the Volve Field. A comparative analysis with benchmark models, such as random forest, long short-term memory networks (LSTM), temporal fusion transformer (TFT), and Kolmogorov–Arnold Networks (KAN) demonstrates the superiority of TKAN. These results confirm that TKAN not only retains the lightweight advantages of KAN but also significantly improves predictive accuracy by incorporating temporal modeling, underscoring its potential for time series prediction in the oil and gas industry.

考虑到资源枯竭和市场波动带来的挑战，准确的石油产量预测已经成为优化油田开发和提高决策过程的关键组成部分。近年来，各种机器学习和深度学习方法被广泛采用。然而，这些方法在准确性和通用性方面仍然存在明显的局限性，通常无法完全捕捉动态油藏环境和多元数据集的复杂性。为了应对这些挑战，我们提出了时间Kolmogorov-Arnold网络（TKAN），这是一种专门为石油生产背景下的多变量时间序列预测设计的新型深度学习架构。TKAN将Kolmogorov-Arnold分解与自适应样条增强激活函数相结合，使模型能够有效地捕获非线性关系和时间依赖性。这使得在处理噪声和动态数据集时，与传统技术相比有了实质性的改进。在实验部分，将提出的TKAN模型应用于Volve油田的产量预测。通过与随机森林、长短期记忆网络（LSTM）、时间融合变压器（TFT）和Kolmogorov-Arnold网络（KAN）等基准模型的对比分析，证明了TKAN的优越性。这些结果证实，TKAN不仅保留了KAN的轻量级优势，而且通过结合时间建模显著提高了预测精度，强调了其在油气行业时间序列预测中的潜力。

{"title":"Oil production forecasting using temporal Kolmogorov–Arnold networks","authors":"Mengze Zheng, Tao Zhang, Jing Cao, Zhong Chen, Jian Zou","doi":"10.1016/j.compchemeng.2025.109483","DOIUrl":"10.1016/j.compchemeng.2025.109483","url":null,"abstract":"<div><div>Given the challenges associated with resource depletion and market volatility, accurate oil production forecasting has become a critical component for optimizing oilfield development and enhancing decision-making processes. In recent years, various machine learning and deep learning methods have been widely adopted. However, these approaches still exhibit significant limitations in terms of accuracy and generalizability, often failing to fully capture the complexities of dynamic reservoir environments and multivariate datasets. To address these challenges, we propose the temporal Kolmogorov–Arnold networks (TKAN), a novel deep learning architecture specifically designed for multivariate time-series forecasting in the context of oil production. TKAN integrates Kolmogorov–Arnold decomposition with adaptive spline-enhanced activation functions, enabling the model to effectively capture nonlinear relationships and temporal dependencies. This leads to substantial improvements over conventional techniques when dealing with noisy and dynamic datasets. In the experimental section, the proposed TKAN model is utilized to predict oil production in the Volve Field. A comparative analysis with benchmark models, such as random forest, long short-term memory networks (LSTM), temporal fusion transformer (TFT), and Kolmogorov–Arnold Networks (KAN) demonstrates the superiority of TKAN. These results confirm that TKAN not only retains the lightweight advantages of KAN but also significantly improves predictive accuracy by incorporating temporal modeling, underscoring its potential for time series prediction in the oil and gas industry.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"205 ","pages":"Article 109483"},"PeriodicalIF":3.9,"publicationDate":"2025-11-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145517214","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","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 framework of intrinsic constraint handling and safe reinforcement learning for crude oil scheduling 原油调度的内在约束处理与安全强化学习混合框架

IF 3.9 2区工程技术 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Chemical Engineering

Pub Date : 2025-11-13 DOI: 10.1016/j.compchemeng.2025.109480

Zhineng Tao , Haoran Li , Tong Qiu

Crude oil scheduling is a critical but highly complex sequential decision-making problem in refinery operations. Traditional mathematical programming methods suffer from exponential computational complexity with increasing scale, while traditional reinforcement learning approaches struggle to guarantee the satisfaction of numerous process and product quality constraints. This highlights a critical issue in the current field of scheduling: the lack of an optimization methodology that can simultaneously achieve high computational efficiency and robust constraint satisfaction. To bridge this gap, we propose a novel hybrid framework based on constraint stratification. The framework embeds critical hard constraints, such as entity connectivity limit, directly into the scheduling environment for intrinsic satisfaction through action masking and shaping. Concurrently, it employs safe reinforcement learning algorithms to manage soft constraints, such as tank inventory levels and product quality specifications, by optimizing a primary objective while keeping constraint violations below a predefined threshold. Through comparative experiments on scheduling cases, the Constrained Policy Optimization algorithm was identified as the most effective safe reinforcement learning method. The results demonstrate that our proposed framework significantly outperforms traditional methods. It achieves the high computational efficiency and scalability of reinforcement learning while providing a much stronger safety guarantee than penalty-based approaches, offering a robust and practical solution for complex industrial scheduling problems.

原油调度是炼油厂生产过程中一个关键而又高度复杂的顺序决策问题。传统的数学规划方法随着规模的增加，计算复杂度呈指数级增长，而传统的强化学习方法难以保证满足众多的过程和产品质量约束。这突出了当前调度领域的一个关键问题：缺乏一种能够同时实现高计算效率和鲁棒约束满足的优化方法。为了弥补这一差距，我们提出了一种基于约束分层的新型混合框架。该框架将实体连接限制等关键硬约束直接嵌入到调度环境中，通过动作遮蔽和塑造实现内在满足。同时，它采用安全的强化学习算法来管理软约束，如坦克库存水平和产品质量规格，通过优化主要目标，同时保持约束违规低于预定义的阈值。通过调度案例对比实验，验证了约束策略优化算法是最有效的安全强化学习方法。结果表明，我们提出的框架显著优于传统方法。它实现了强化学习的高计算效率和可扩展性，同时提供了比基于惩罚的方法更强的安全性保证，为复杂的工业调度问题提供了鲁棒性和实用性的解决方案。

{"title":"A hybrid framework of intrinsic constraint handling and safe reinforcement learning for crude oil scheduling","authors":"Zhineng Tao , Haoran Li , Tong Qiu","doi":"10.1016/j.compchemeng.2025.109480","DOIUrl":"10.1016/j.compchemeng.2025.109480","url":null,"abstract":"<div><div>Crude oil scheduling is a critical but highly complex sequential decision-making problem in refinery operations. Traditional mathematical programming methods suffer from exponential computational complexity with increasing scale, while traditional reinforcement learning approaches struggle to guarantee the satisfaction of numerous process and product quality constraints. This highlights a critical issue in the current field of scheduling: the lack of an optimization methodology that can simultaneously achieve high computational efficiency and robust constraint satisfaction. To bridge this gap, we propose a novel hybrid framework based on constraint stratification. The framework embeds critical hard constraints, such as entity connectivity limit, directly into the scheduling environment for intrinsic satisfaction through action masking and shaping. Concurrently, it employs safe reinforcement learning algorithms to manage soft constraints, such as tank inventory levels and product quality specifications, by optimizing a primary objective while keeping constraint violations below a predefined threshold. Through comparative experiments on scheduling cases, the Constrained Policy Optimization algorithm was identified as the most effective safe reinforcement learning method. The results demonstrate that our proposed framework significantly outperforms traditional methods. It achieves the high computational efficiency and scalability of reinforcement learning while providing a much stronger safety guarantee than penalty-based approaches, offering a robust and practical solution for complex industrial scheduling problems.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"205 ","pages":"Article 109480"},"PeriodicalIF":3.9,"publicationDate":"2025-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145576348","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

An optimization framework for renewable liquefied gas supply chain: A case study on glycerol-based production 可再生液化气供应链的优化框架：以甘油为基础的生产为例

IF 3.9 2区工程技术 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Chemical Engineering

Pub Date : 2025-11-13 DOI: 10.1016/j.compchemeng.2025.109481

Larissa Thaís Bruschi, Luiz Kulay, Moisés Teles dos Santos

Renewable Liquefied Gas (RLG) is emerging as a viable alternative to Liquefied Petroleum Gas (LPG), providing environmental benefits and strengthening energy security. However, the implementation of biofuels faces challenges such as long-term feedstock availability, technology scalability, facility site selection, and tax policies. This study proposes a Mixed Integer Linear Programming optimization model to support strategic and tactical RLG supply chain decisions in Brazil. The model determines the optimal feedstock purchasing, facility locations, processing capacities, material flows, and product distribution to maximize the Net Present Value (NPV), in a case study that explores the production of RLG from glycerol. For the economic feasibility of the supply chain the RLG selling price must be 80 % higher than that of LPG. The optimal network favors a centralized system, where transportation costs minimally impact NPV, while economies of scale lower investment and production costs. A decentralized approach is evaluated by limiting the installed production capacity. Although a decentralized network requires an RLG selling price 220 % higher than LPG, it allows smaller stakeholders to enter the market and achieves the same payback period as the centralized system. Sensitivity analysis on conversion yield shows that a 5 % efficiency increase shortens the payback period by four years and triples the NPV. The study highlights the importance of supply chain integration, cost reduction strategies, and policy incentives in enhancing RLG competitiveness, thereby contributing to its deployment despite facing economic and logistical challenges.

可再生液化气（RLG）正在成为液化石油气（LPG）的可行替代品，具有环保效益，并加强了能源安全。然而，生物燃料的实施面临着诸如长期原料可用性、技术可扩展性、设施选址和税收政策等挑战。本研究提出一个混合整数线性规划优化模型，以支持巴西RLG供应链的战略和战术决策。该模型确定了最佳的原料采购、设施位置、处理能力、物料流和产品分配，以最大化净现值（NPV），在一个探索从甘油中生产RLG的案例研究中。为了供应链的经济可行性，RLG的销售价格必须比LPG高80%。最优网络倾向于集中式系统，在这种系统中，运输成本对净现值的影响最小，而规模经济则降低了投资和生产成本。通过限制已安装的生产能力来评估分散式方法。尽管分散式网络要求RLG的销售价格比LPG高220%，但它允许较小的利益相关者进入市场，并实现与集中式系统相同的投资回收期。对转化率的敏感性分析表明，效率提高5%，投资回收期缩短4年，净现值增加3倍。该研究强调了供应链整合、降低成本战略和政策激励在提高RLG竞争力方面的重要性，从而有助于在面临经济和物流挑战的情况下部署RLG。

{"title":"An optimization framework for renewable liquefied gas supply chain: A case study on glycerol-based production","authors":"Larissa Thaís Bruschi, Luiz Kulay, Moisés Teles dos Santos","doi":"10.1016/j.compchemeng.2025.109481","DOIUrl":"10.1016/j.compchemeng.2025.109481","url":null,"abstract":"<div><div>Renewable Liquefied Gas (RLG) is emerging as a viable alternative to Liquefied Petroleum Gas (LPG), providing environmental benefits and strengthening energy security. However, the implementation of biofuels faces challenges such as long-term feedstock availability, technology scalability, facility site selection, and tax policies. This study proposes a Mixed Integer Linear Programming optimization model to support strategic and tactical RLG supply chain decisions in Brazil. The model determines the optimal feedstock purchasing, facility locations, processing capacities, material flows, and product distribution to maximize the Net Present Value (NPV), in a case study that explores the production of RLG from glycerol. For the economic feasibility of the supply chain the RLG selling price must be 80 % higher than that of LPG. The optimal network favors a centralized system, where transportation costs minimally impact NPV, while economies of scale lower investment and production costs. A decentralized approach is evaluated by limiting the installed production capacity. Although a decentralized network requires an RLG selling price 220 % higher than LPG, it allows smaller stakeholders to enter the market and achieves the same payback period as the centralized system. Sensitivity analysis on conversion yield shows that a 5 % efficiency increase shortens the payback period by four years and triples the NPV. The study highlights the importance of supply chain integration, cost reduction strategies, and policy incentives in enhancing RLG competitiveness, thereby contributing to its deployment despite facing economic and logistical challenges.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"205 ","pages":"Article 109481"},"PeriodicalIF":3.9,"publicationDate":"2025-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145526245","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A correction symbol Granger causality approach to root cause diagnosis of non-stationary industrial processes 非平稳工业过程根本原因诊断的校正符号格兰杰因果关系方法

IF 3.9 2区工程技术 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Chemical Engineering

Pub Date : 2025-11-12 DOI: 10.1016/j.compchemeng.2025.109479

Rui Chen , Husnain Ali , Shu Liang , Yuanqiang Zhou , Furong Gao

Dealing with non-stationary characteristics and revealing the symbol properties of causalities remain obstacles to performing fault root cause diagnosis (RCD) in industrial processes. Existing methods overlook these problems, leading to reduced RCD accuracy and a limited understanding of the dynamic characteristics of fault propagation. To address the above-mentioned problems, this paper proposes a correction symbol Granger causality (CSGC) approach. Firstly, to infer significant Granger causality (GC) from non-stationary processes, the cointegration relationships among variables are utilized to differentiate them and construct an error correction term, enabling the regression model to capture the long-term dependencies among fault variables better. Secondly, the self-explaining neural network is employed to further reveal the symbol properties of the fault variable from the fitted generalized coefficient matrix. Furthermore, two causal criteria of CSGC are established to test the significance level of GC and eliminate bidirectional causalities. Based on CSGC, a causal graph construction strategy containing information on symbol properties is developed for fault RCD and analysis of fault propagation paths. The analysis results of the numerical simulation system indicate that the various metrics of CSGC perform well, demonstrating its potential for accurate RCD and in-depth fault propagation analysis (Accuracy: 0.96, Sensitivity: 1.00, Specificity: 0.95, F1 score: 0.91). Finally, the effectiveness of RCD is validated through its application to the numerical simulation, the Tennessee Eastman process (TEP) and a real-world motor soft-foot fault.

处理非平稳特征和揭示因果关系的符号属性仍然是进行工业过程故障根本原因诊断（RCD）的障碍。现有方法忽略了这些问题，导致RCD精度降低，对故障传播动态特性的理解有限。针对上述问题，本文提出了一种修正符号格兰杰因果关系（CSGC）方法。首先，为了从非平稳过程中推断出显著的格兰杰因果关系（GC），利用变量间的协整关系进行区分并构造误差修正项，使回归模型能够更好地捕捉故障变量之间的长期依赖关系。其次，利用自解释神经网络从拟合的广义系数矩阵中进一步揭示故障变量的符号性质；此外，建立了CSGC的两个因果标准，检验了GC的显著性水平，消除了双向因果关系。基于CSGC，提出了一种包含符号属性信息的因果图构建策略，用于故障RCD和故障传播路径分析。数值模拟系统的分析结果表明，CSGC的各项指标表现良好，显示了其在精确RCD和深度故障传播分析方面的潜力（精度：0.96，灵敏度：1.00，特异性：0.95，F1评分：0.91）。最后，通过数值模拟、田纳西伊士曼过程（Tennessee Eastman process， TEP）和实际电机软足故障实例验证了RCD方法的有效性。

{"title":"A correction symbol Granger causality approach to root cause diagnosis of non-stationary industrial processes","authors":"Rui Chen , Husnain Ali , Shu Liang , Yuanqiang Zhou , Furong Gao","doi":"10.1016/j.compchemeng.2025.109479","DOIUrl":"10.1016/j.compchemeng.2025.109479","url":null,"abstract":"<div><div>Dealing with non-stationary characteristics and revealing the symbol properties of causalities remain obstacles to performing fault root cause diagnosis (RCD) in industrial processes. Existing methods overlook these problems, leading to reduced RCD accuracy and a limited understanding of the dynamic characteristics of fault propagation. To address the above-mentioned problems, this paper proposes a correction symbol Granger causality (CSGC) approach. Firstly, to infer significant Granger causality (GC) from non-stationary processes, the cointegration relationships among variables are utilized to differentiate them and construct an error correction term, enabling the regression model to capture the long-term dependencies among fault variables better. Secondly, the self-explaining neural network is employed to further reveal the symbol properties of the fault variable from the fitted generalized coefficient matrix. Furthermore, two causal criteria of CSGC are established to test the significance level of GC and eliminate bidirectional causalities. Based on CSGC, a causal graph construction strategy containing information on symbol properties is developed for fault RCD and analysis of fault propagation paths. The analysis results of the numerical simulation system indicate that the various metrics of CSGC perform well, demonstrating its potential for accurate RCD and in-depth fault propagation analysis (Accuracy: 0.96, Sensitivity: 1.00, Specificity: 0.95, F1 score: 0.91). Finally, the effectiveness of RCD is validated through its application to the numerical simulation, the Tennessee Eastman process (TEP) and a real-world motor soft-foot fault.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"205 ","pages":"Article 109479"},"PeriodicalIF":3.9,"publicationDate":"2025-11-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145517220","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

L-TKAN:A fast and accurate Laplacian radial basis function-Based Temporal Kolmogorov–Arnold Network for state of charge estimation of lithium-ion batteries L-TKAN：一种快速准确的基于拉普拉斯径向基函数的锂离子电池充电状态估计时态Kolmogorov-Arnold网络

IF 3.9 2区工程技术 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Chemical Engineering

Pub Date : 2025-11-10 DOI: 10.1016/j.compchemeng.2025.109478

Zhiqiang Liu , Chong Kuai , Gang Wu , Ashun Zang

Traditional SOC estimation methods are susceptible to temperature variations and external disturbances, while conventional data-driven methods can effectively deal with these issues but struggle with nonlinear issues. The Kolmogorov–Arnold Network (KAN), a novel network configuration, demonstrates excellent performance in handling nonlinear problems. Nevertheless, the use of B-spline functions in KAN results in slow training speed. This paper proposes a novel neural architecture network, L-TKAN, by replacing the B-spline function (RMSE1.41%,MAE1.12%,R²0.997444) with the Laplacian radial basis function (Laplacian RBF) (RMSE1.25%,MAE0.92%,R²0.997988) to address this issue. The corresponding 300-epoch training times of these two configurations were 288 min and 375 min, respectively. The results demonstrate that the use of the Laplacian RBF not only significantly accelerates the training speed but also improves the model performance. Moreover, compared to the traditional Gaussian RBF, the Laplacian RBF also exhibits these two advantages (RMSE 1.35% and 1.39%; MAE 1% and 1.04%; R² 0.997657 and 0.997524, the shortest training time of 288 min and 294 min, respectively). During the experiments, we discovered the relationship between two key parameters (the number of center points and the decay factor

σ

) of the Laplacian RBF. Based on this finding, we further improved the performance of the model-When the number of center points was 2, the RMSE decreased 0.06%. Furthermore, the best performance was achieved with 1 center point, yielding an RMSE of 1.25%, MAE of 0.92%, and R² of 0.997988. The proposed model also outperforms other models, such as KAN, MLP, LSTM, GRU, TCN, CNNLSTM, CNNGRU, transformer, CNNtransformer.

传统的SOC估计方法容易受到温度变化和外界干扰的影响，而传统的数据驱动方法可以有效地处理这些问题，但难以解决非线性问题。Kolmogorov-Arnold网络（KAN）是一种新颖的网络结构，在处理非线性问题方面表现出优异的性能。然而，在KAN中使用b样条函数导致训练速度慢。本文采用拉普拉斯径向基函数（Laplacian RBF）（RMSE1.25%,MAE0.92%,R20.997988）代替b样条函数（RMSE1.41%,MAE1.12%,R20.997444），提出了一种新的神经网络结构L-TKAN。这两种配置对应的300 epoch训练时间分别为288 min和375 min。结果表明，使用拉普拉斯RBF不仅可以显著提高训练速度，而且可以提高模型的性能。此外，与传统的高斯RBF相比，拉普拉斯RBF也表现出这两个优势（RMSE分别为1.35%和1.39%；MAE分别为1%和1.04%；R2分别为0.997657和0.997524，最短的训练时间分别为288 min和294 min）。在实验中，我们发现了拉普拉斯RBF的两个关键参数（中心点个数和衰减因子σ）之间的关系。基于这一发现，我们进一步改进了模型的性能——当中心点个数为2时，RMSE下降了0.06%。以1个中心点为最佳，RMSE为1.25%，MAE为0.92%，R2为0.997988。该模型也优于其他模型，如KAN、MLP、LSTM、GRU、TCN、CNNLSTM、CNNGRU、transformer、CNNtransformer。

{"title":"L-TKAN:A fast and accurate Laplacian radial basis function-Based Temporal Kolmogorov–Arnold Network for state of charge estimation of lithium-ion batteries","authors":"Zhiqiang Liu , Chong Kuai , Gang Wu , Ashun Zang","doi":"10.1016/j.compchemeng.2025.109478","DOIUrl":"10.1016/j.compchemeng.2025.109478","url":null,"abstract":"<div><div>Traditional SOC estimation methods are susceptible to temperature variations and external disturbances, while conventional data-driven methods can effectively deal with these issues but struggle with nonlinear issues. The Kolmogorov–Arnold Network (KAN), a novel network configuration, demonstrates excellent performance in handling nonlinear problems. Nevertheless, the use of B-spline functions in KAN results in slow training speed. This paper proposes a novel neural architecture network, L-TKAN, by replacing the B-spline function (RMSE1.41%,MAE1.12%,R<sup>2</sup>0.997444) with the Laplacian radial basis function (Laplacian RBF) (RMSE1.25%,MAE0.92%,R<sup>2</sup>0.997988) to address this issue. The corresponding 300-epoch training times of these two configurations were 288 min and 375 min, respectively. The results demonstrate that the use of the Laplacian RBF not only significantly accelerates the training speed but also improves the model performance. Moreover, compared to the traditional Gaussian RBF, the Laplacian RBF also exhibits these two advantages (RMSE 1.35% and 1.39%; MAE 1% and 1.04%; R<sup>2</sup> 0.997657 and 0.997524, the shortest training time of 288 min and 294 min, respectively). During the experiments, we discovered the relationship between two key parameters (the number of center points and the decay factor <span><math><mi>σ</mi></math></span>) of the Laplacian RBF. Based on this finding, we further improved the performance of the model-When the number of center points was 2, the RMSE decreased 0.06%. Furthermore, the best performance was achieved with 1 center point, yielding an RMSE of 1.25%, MAE of 0.92%, and R<sup>2</sup> of 0.997988. The proposed model also outperforms other models, such as KAN, MLP, LSTM, GRU, TCN, CNNLSTM, CNNGRU, transformer, CNNtransformer.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"205 ","pages":"Article 109478"},"PeriodicalIF":3.9,"publicationDate":"2025-11-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145517223","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0