This investigation presents a robust spectral technique for accurately solving timespace fractional advection-diffusion equations (FADEs) characterized by Caputo-type derivatives with coefficients that vary in both time and space. Such equations inherently model nonlocality and memory effects prevalent in various complex transport and physical systems. The proposed strategy reformulates the underlying problem as a system of fractional-order ordinary differential equations (FODEs), leveraging operational matrices derived from shifted Jacobi polynomials (SJPs) within the spatial discretization framework. The initial conditions of the resulting FODEs are directly inherited from those of the original equation. The construction of explicit particular solutions for each FODE relies on the introduction of auxiliary initial value problems. Subsequently, the particular solutions are assembled into a linear expansion designed to minimize the residual error across the domain. We employ a weighted residual procedure to perform the minimization, ensuring average error suppression and greater solution precision. Unlike conventional spectral operational matrix methods that rely on full discretization, the present approach constructs explicit particular solutions for the resulting FODEs and determines the temporal coefficients through a residual minimization procedure. A rigorous theoretical validation is conducted via residual-based error estimation, confirming the convergence behavior of the scheme. Performance is assessed using test problems, where empirical convergence rates are calculated and compared with those obtained via other numerical methods. The comparative analysis underscores the enhanced accuracy and robustness of the method under consideration. These results demonstrate the method’s potential as a reliable tool for solving FADEs in scientific and mathematical applications.
The present work shows the numerical performances through stochastic paradigms of the reactive transport system (RTS). The designed construction of the solver is presented through the Morlet wavelet function, which works as an artificial neural network and the hybridization of genetic algorithm and active set. An objective function is created using the differential RTS, which is optimized through the hybrid capability of the designed scheme. The precision of the solver is perceived in comparison with the obtained and reference outcomes, absolute error, and number of multiple executions based on different operators. The inspiration to exploit the Morlet wavelet neural network process and the hybridization methods comes to present precise and reliable solutions for solving the nonlinear and stiff differential systems. Three RTS cases are numerically treated to validate the efficiency of the designed process, and the performance of the scheme is also validated by using different statistical operators.
Green credit risk assessment for manufacturing enterprises is a dynamic and complex problem affected by operational states, environmental uncertainty, financial conditions, information quality, and interactive risk evolution. Traditional assessment methods depend heavily on low-frequency financial and disclosure data and therefore have limited ability to capture early risk signals arising from production operations, energy consumption, environmental performance, and supply-chain disturbances. In addition, conventional machine-learning models provide limited interpretability for causal dependence and uncertain risk propagation. This paper proposes a digital twin-driven hybrid fuzzy dynamic Bayesian network (DT-FDBN) framework for dynamic green credit risk assessment and scenario simulation in manufacturing enterprises. The framework integrates multi-source information covering financial solvency, production and equipment operation, energy consumption and emissions, green transition, environmental disclosure and information credibility, and the external environment into a unified enterprise digital-twin representation. Fuzzy evidence transformation is used to represent incomplete and imprecise observations, while credibility correction and group-level sensor reliability reduce the influence of unreliable digital-twin evidence. The DT-FDBN further models hierarchical risk dependence, temporal evolution, and probabilistic uncertainty propagation, supporting dynamic inference, key-factor diagnosis, multi-period prediction, and intervention-scenario simulation. Based on a semi-synthetic dataset containing 2,160 enterprise-month observations from 60 manufacturing enterprises, the proposed framework achieves an accuracy of 0.881, recall of 0.893, F1-score of 0.869, AUC of 0.943, and Brier score of 0.079. It outperforms XGBoost and conventional fuzzy dynamic Bayesian networks in overall predictive and calibration performance and identifies operational and environmental deterioration earlier than financial-only assessment. The proposed framework provides an interpretable and dynamically adaptive modeling paradigm for manufacturing green credit risk management.
This study addresses the heterogeneity of multi-source human resource data, the ambiguity of subjective evaluations, the time-varying nature of performance states, and the difficulty of translating predictions into management interventions. A hybrid fuzzy-dynamic Bayesian performance management framework, HFDB-PM, is proposed. The framework aligns employee attributes, the management environment, cognitive-affective states, behavioral responses, and performance outcomes; represents expert evaluations using triangular fuzzy numbers; and integrates theoretical constraints, historical data, and expert knowledge to learn a dynamic Bayesian network. Online updating, backward inference, sensitivity analysis, and multi-period Monte Carlo simulation are further combined to support performance prediction, causal diagnosis, and intervention optimization. Experiments are conducted on a semi-synthetic longitudinal dataset containing 1,284 employees, eight quarters, and 10,272 employee-time-slice records. HFDB-PM achieves an Accuracy of 0.856, a Macro-F1 of 0.838, and an AUC of 0.925, with a Brier Score of 0.098 and an ECE of 0.021. Compared with DBN, Macro-F1 and AUC increase by 3.1 and 2.1 percentage points, respectively, while ECE decreases by 40.0%. In the three-period simulation, the combined strategy of training, transparent feedback, and workload adjustment raises the probability of high performance to 0.468 and reduces the probability of low performance to 0.223, demonstrating the effectiveness of the framework for dynamic prediction, interpretable diagnosis, and management decision support.
By integrating blockchain technology with data-driven intelligent decision-making, the framework ensures trustworthy supply chain data management and procurement optimization. A multi-channel consortium blockchain architecture enables secure data sharing among suppliers, manufacturers, distributors, and end-users. Smart contracts automate transaction verification and recording, ensuring data authenticity, immutability, and full lifecycle traceability This paper proposes a blockchain-based traceability framework to support the green transformation of optoelectronic enterprises through secure, transparent, and intelligent supply network operations. An on-chain/off-chain collaborative storage mechanism reduces storage overhead while preserving data integrity. Furthermore, a reinforcement learning-based procurement decision model combines historical production data and market intelligence to optimize raw material purchasing strategies, inventory management, and cost control. Experimental results show that traceability query time is reduced to approximately 95 ms, improving efficiency by over 60% compared with traditional approaches. The system achieves a tamper detection rate above 99%, reduces inventory holding costs by about 20%, and lowers total procurement costs by approximately 10%, demonstrating significant benefits for sustainable supply chain management.
The growing complexity of computer viruses has necessitated the development of more sophisticated mathematical models to understand and control cyber-epidemics. Building on classical compartmental approaches, we extend the Susceptible-Infected-Removed-Antidotal (SIRA) model into a reactiondiffusion framework to capture both local infection dynamics and spatial transmission across networks. The resulting system presents significant analytical challenges due to stiffness, positivity constraints, and nonlinear couplings. To address these issues, we propose a semi-implicit numerical scheme that treats diffusion terms implicitly for stability and reaction terms explicitly for computational efficiency, coupled with Nonstandard Finite Difference (NSFD) techniques to ensure solution positivity. Rigorous stability analysis based on M-matrix theory establishes conditions for boundedness and convergence of the discrete system. Compared to existing methodologies, our approach enables accurate, efficient simulation of large-scale malware propagation, offering critical insights for cybersecurity strategies such as antivirus deployment optimization and outbreak containment. Numerical experiments demonstrate the effectiveness of the proposed scheme in capturing complex spatiotemporal dynamics of cyber threats.
Despite offering scalable solutions through advanced cloud applications in healthcare, cloud computing poses additional risks in cybersecurity that need to be taken seriously. With the introduction of the Internet of Medical Things (IoMT), the requirement for adaptive and advanced intrusion detection systems has become more pronounced. Intrusion detection methods are currently inefficient in dealing with the multidimensional complexity of information generated by IoMT devices in real time. It is currently difficult to achieve both high accuracy rates and speed when detecting attacks in current approaches. To meet these demands, there is an increased need for optimization of adaptive detection methods. The current study introduces Transforming Cloud-based Healthcare Security with Parrot Optimized Deep Graph Network Propagation for Intrusion Detection Systems (ADGNP–PIO), an innovative hybrid intrusion detection system that focuses on enhancing the effectiveness and efficiency of the detection process in cloud-based healthcare systems. Based on the CICIoMT2024 and WUSTL-EHMS-2020 datasets, Adaptive Deep Graph Network Propagation with Parrot Insight Optimizer (ADGNP–PIO) uses the Max–Min Scaling Normalizer (MaMiSN) to pre-process data by eliminating redundancies and making sure features are normalized in terms of their ranges before being used in the feature selection phase. In order to choose optimal features, the Synergistic Squid–Penguin Hybrid Optimizer (SSq–PHO) will be employed to identify important data features. By making use of relationships between features in the IoT data, the Adaptive Deep Graph Network Propagation (ADGNP) will detect complex attacks and reduce false-positive results. Moreover, using the Parrot Insight Optimizer (PIO), the network parameters are optimized to provide optimal performance when dealing with IoMT data streams. Indeed, the accuracy of this algorithm on the WUSTL-EHMS-2020 dataset is excellent with values of 99.45% accuracy, 99.32% precision, 99.51% recall, and 99.41% [Formula: see text]1-score; as well as with the CICIoMT2024 dataset having accuracy values of 99.52%, precision of 99.48%, recall of 99.57%, and [Formula: see text]1-score of 99.50%. These observations prove the performance and scalability of ADGNP–PIO, reinforcing its suitability as a robust intrusion detection solution for highly specialized medical cloud networks.
Environmental, social, and governance (ESG) performance has become a key dimension for assessing corporate sustainable competitiveness and long-term value, yet existing methods have difficulty simultaneously addressing multi-source heterogeneous information, fuzzy assessments, intertemporal evolution, and multi-constraint optimization. To address these issues, this study proposes a distributed dynamic hybrid fuzzy–Bayesian decision model for AI-driven ESG governance. The model comprises six modules—AI capability, action mechanisms, environmental performance, social performance, governance performance, and strategic performance. Fuzzy soft evidence and belief rules are used to integrate continuous, discrete, and linguistic information; a dual-population cooperative genetic algorithm learns the network structure under theoretical constraints; and dynamic inference, risk-contribution diagnosis, and simulated annealing are combined to optimize corporate strategies. Under the tested 2012-2024 Chinese A-share sample and the designed simulation scenarios, the model achieved an Accuracy of 0.853, a Macro-F1 of 0.838, and an AUC of 0.912, while reducing Log Loss and the Brier Score to 0.441 and 0.108, respectively. Under the model-optimized scenario, ESG risk decreased from 0.346 to 0.186, a reduction of 46.2%, while the ESG coordination index and comprehensive utility reached 0.718 and 0.759. Within this empirical and simulation setting, the results indicate that the proposed method improves ESG state recognition, risk-probability calibration, and strategic resource allocation under a limited budget.
Purpose The goal of this research is to create a framework for financial system fraud detection that is safe and protects privacy while resolving issues with data sharing, legal restrictions, and cybersecurity threats. To facilitate cooperative model training without exchanging private raw data. Methods A blockchain-enhanced Federated Learning (FL) method is suggested. The technique divides a centralized dataset among several clients to simulate an FL environment. Local training is done using a TabNet deep learning model, and hyperparameter tweaking is done via Bayesian optimization. Input features are normalized and passed through TabNet's sequential feature transformer and attentive transformer, which generate sparse attention masks to select the most discriminative attributes at each decision step; the aggregated decision representations are then classified using a softmax-based prediction layer. Model updates are transmitted using AES-256 encryption to guarantee data security, and their integrity is confirmed by SHA-256 hashing. Blockchain technology provides transparency and tamper resistance by recording model update hashes via Hyperledger Fabric. The integration of AES-256 encryption, SHA-256 integrity verification, and Hyperledger Fabric blockchain enables secure peer-to-peer model transmission, making the proposed framework well suited for decentralized financial environments. Results With an accuracy of 0.9924, precision of 0.9947, recall of 0.9910, and low rates of false positives and false negatives, the findings show excellent performance. Validation accuracy improved consistently across successive federated communication rounds, rising from 0.9887 to 0.9933, confirming stable model convergence, while the framework achieved a ROC-AUC of 0.9980 and PR-AUC of 0.9986. The suggested architecture performs better than conventional models like decision trees and logistic regression. Conclusion The framework is appropriate for practical financial applications since it successfully improves fraud detection while guaranteeing Privacy Preservation, security, and openness.
Partially observed epidemic systems are difficult to analyze because confirmed cases provide only an indirect view of transmission and hidden infections are usually unobserved. We consider the susceptible–unidentified infected–confirmed (SUC) epidemic model and develop SUC–PINN, a physics-informed neural network framework for estimating hidden epidemic states, transmission and confirmation rates, and the initial unidentified population from confirmed-case time series alone. The method combines confirmed-case observations with SUC residuals and initial-condition constraints, so that the learned trajectories remain tied to the governing dynamics. Experiments with synthetic data and COVID-19 surveillance data from Indonesia, Thailand, India, and the Philippines show that SUC–PINN recovers plausible hidden infection trajectories, yields stable parameter estimates, and provides accurate short-term forecasts. These results support SUC–PINN as a practical computational approach for inverse modeling and prediction in partially observed epidemic dynamics.
New quality productivity (NQP) is an important force for economic and social progress, and in the industrial sector—the core of the real economy—its transformation bears directly on national competitiveness and high-quality, sustainable growth. Focusing on the R&D-innovation core of industrial NQP, this paper constructs an evaluation system of seven indicators along the “R&D investment–activity–output” chain for 31 Chinese provinces over 2015–2023. Using the entropy method and grey relational analysis, it identifies each province's core driving indicator and proposes the concept of “heterogeneity of core driving indicators”; a PSO-optimized fractional grey multivariate model, FGM(1,2), is then used to forecast NQP for 2024–2028, outperforming ARIMA, conventional grey models and machine-learning benchmarks. The results reveal an overall upward but regionally uneven trend: the East advances rapidly and is primarily R&D-investment-driven, the Central region grows fastest, the Northeast lags, and the West achieves localized, policy-supported breakthroughs with output-driven features. These findings support differentiated, region-specific strategies for sustained NQP growth and high-quality regional development.
While intelligently evaluating martial arts actions, we often face multiple uncertainties, such as video quality degradation, partial occlusion, and subjective evaluation fuzziness. Thus, the robustness and flexibility of the evaluation system are always restricted. To address these challenges, this study proposes a hybrid fuzzy Bayesian resilient simulation framework for dynamic complex systems. Through probabilistic modeling of the system state space, the framework transforms the action quality evaluation into the recursive estimation of the posterior confidence of the state, that is, the online inference of a dynamic Bayesian network is adopted to simulate the behavior trajectory of the system driven by uncertainty. Firstly, an improved double convolutional neural network is used to extract spatio-temporal features from video sequences accurately. Then, a fuzzy semantic quantification module is designed to map the quantitative features into qualitative evaluation grades to deal with the semantic fuzziness in the evaluation process. Finally, by using the previous state prior and the current fuzzy evidence, a dynamic Bayesian network is constructed to recursively infer the posterior state confidence with time continuity, which intends to maintain the stability of system inference when information is missing or contaminated. To further verify the flexibility of the framework, this study not only evaluates its basic recognition accuracy on the ActivityNet and THU-MOS14 datasets, but also designs the interference scenarios, including simulated physical occlusion and semantic boundary tampering. Besides, it also systematically tests the uncertainty tolerance and resilience of the model. The experimental findings show that the hybrid framework can not only maintain the recognition accuracy comparable to the advanced CNN baseline model, but also reduce the maximum attenuation of confidence by about 41.6% under simulated occlusion interference, and shorten the elastic recovery time by more than 50%. Thus, this study provides a theoretical basis and practical framework for the construction of a highly robust cyber physical simulation system for intangible cultural heritage protection and intelligent physical education teaching.
Intelligent transportation systems depend on accurately detecting and tracking objects, such as pedestrians, vehicles, etc., from driving footage to make effective decisions. The conventional schemes rely on data collected from multiple vehicles for Multi-Object Detection and Tracking (MODT), which poses a great risk to user privacy and security. Federated Learning (FL) facilitates collaborative model training across distributed devices while maintaining data privacy. The existing techniques often suffer from generalization issues and low convergence. Hence, an efficient model, namely Chronological Electrical Eel Foraging Optimization Algorithm_You Only Look Once version 12 (Cr-EEFO_YOLOv12), is proposed for MODT in FL. In this approach, the Cr-EEFO metaheuristic is employed to optimize the parameters of YOLOv12, improving detection performance and convergence during distributed training. In the FL framework, various entities, like nodes and servers, are involved. The local training is performed by considering local data. The server updation is performed by data aggregation, and the global model is downloaded to the nodes. During training, the input video is subjected to frame extraction. Then, object segmentation is accomplished utilizing Modified Fuzzy C-means Clustering (MFCM). Afterwards, multiple object detection is done using YOLOv12, trained using Cr-EEFO. Lastly, multi-object tracking is done using an Unscented Kalman Filter (UKF). Cr-EEFO_YOLOv12 attained a True Positive Rate (TPR), Mean Square Error (MSE), accuracy, True Negative Rate (TNR), and mean Average Precision (mAP) of 97.568%, 0.067, 96.667%, 96.358%, and 83.889% indicating its effectiveness in improving detection reliability and learning performance compared to existing approaches, when evaluated on the Driving Video with Object Tracking dataset.
Process modeling and simulation (M&S) is a well-established paradigm for analyzing and designing complex systems. While commercial tools like SimEvents and AnyLogic offer powerful, component-based modeling capabilities, their underlying mechanisms are often opaque, limiting deep extension and customization for advanced research. In our work, we address this critical gap by proposing a formal, three-part building-block methodology for constructing discrete process M&S components. Our approach, grounded in and extending the discrete event system specification (DEVS) formalism, decomposes component design into three distinct yet interconnected facets: the internal hierarchical structure ([Formula: see text]-Facet) of individual components; the nesting and composition relationships ([Formula: see text]-Facet) between components; and a novel, formalized inter-component collaboration protocol ([Formula: see text]-Facet) termed the entity flow-control flow-processing logic (ECPR) strategy. This methodology is engineered explicitly for transparency and extensibility, bridging the gap between the usability of commercial tools and the flexibility of open formalisms. We present its implementation within the open X-Language/XLAB framework, and validate it through three comprehensive case studies: a vehicle CAN bus/ABS system, a computational graph executor, and a custom component extension. Results demonstrate not only functional equivalence with commercial tools (Simulink, AnyLogic), but also, through a comparative architectural analysis and a usability pilot study, showcase superior extensibility and a manageable learning curve. This work provides a reference architecture for open, component-based simulation, enabling both intuitive graphical modeling and researcher-accessible simulation kernel extension.
Importance of the study: This study contributes to the implementation of the Physics-Informed Neural Network (PINN) approach for fluid models and ensures physically consistent solutions for temperature and velocity fields. Aim of the study: This paper aims to implement the PINN for an established fluid model to explore the nature of natural convection and heat transfer of a ternary nanofluid in a rectangular cavity under complex thermal boundary conditions. This study examines the interaction of heat sources, sinks, viscous dissipation, and magnetic fields on thermal performance. Physical parameters are used to assess their impact on streamline patterns, isotherm contours, and the average Nusselt number. Causal Physics-Informed Neural Networks (cPINNs) Methodology: The authors employ the model using four hidden layers and 32 neurons in the cPINN approach to solve the thermal engineering model problem involving the ternary nanofluid. The authors employ the preconditioned optimizer PDFP to implement the cPINNs. Vital key outcomes: The results are significant in the cooling of electronic devices, energy storage devices, and solar power, where the regulation of temperature and the efficiency of heat transfer are important. The graphical abstract summarizes the integrated workflow of the proposed PINN-based thermal engineering study, highlighting the connection between mathematical modeling, physics-informed learning, and numerical validation. It provides a concise visual overview of the methodology and multifactorial analysis employed in this work.
This paper presents a synergistic control strategy for discrete-time singularly perturbed systems, where data-driven learning is seamlessly combined with LMI-based synthesis. The proposed method fully leverages data collected from system operation or experiments to establish the input–output relationship for controller parameter identification, thereby enabling controller design without requiring an exact mathematical model of the system. First, the influence mechanism of the singular perturbation parameter on system dynamics is analyzed, which reveals the coupling between the fast and slow subsystems and their respective effects on system stability and response speed. Second, without requiring a complete parametric model of the system, valid matrix inequality conditions are constructed based on input–output data for both controller synthesis and singular perturbation parameter estimation. This approach effectively avoids over-reliance on an accurate system model, thus enhancing the feasibility and practical applicability of the proposed method. Lastly, numerous simulation runs are executed to assess the performance of the presented strategy. The results demonstrate that the method not only ensures the global stability of the system across a wide range of singular perturbation parameter values but also achieves excellent performance in terms of convergence speed, robustness and control accuracy, thus offering an effective new approach for controlling discrete-time singularly perturbed systems in complex engineering environments.
This paper presents a spatiotemporal mathematical model to study the dynamics of the immune system and its response to virus infection during treatment. The model explores the temporal dynamics of the system, performs a stability analysis of the equilibrium points, and investigates instability in the spatiotemporal system using Turing bifurcation conditions. Numerical simulations are conducted to validate the theoretical results and thoroughly investigate the Turing instability boundaries in the parameter space. Induced pattern formations are simulated, and the effects of key parameter variations are examined. Overall, this paper contributes to the understanding of immunosuppressive infection dynamics and provides insights into the behavior of the immune system during treatment.
Heritage conservation of traditional Chinese timber-framed architecture faces a dual challenge: The imminent loss of craftsmanship knowledge transmitted through master–apprentice traditions, and the progressive degradation and deformation of wooden components caused by environmental exposure and repeated restoration interventions. Among its key structural elements, the wing corner is one of the most iconic assemblies in traditional timber architecture. It exhibits complex geometric characteristics, including bi-directional curvature, intricate mortise-and-tenon joints, and significant regional variations between northern and southern building traditions. Existing Scan-to-BIM approaches can only generate noneditable static mesh models, which are insufficient for renovation simulation and structural performance assessment. Parametric modeling techniques, which encode construction rules from historical treatises into computational representations, provide a promising alternative. This study proposes a component-level parametric framework implemented in Dynamo, establishing a parameter-driven four-stage modeling workflow to generate two typologically distinct wing corner forms: The rigid northern official style governed by the Yingzao Fashi, and the flexible southern garden style governed by the Yingzao Fayuan. Validation using case studies of the Hall of Supreme Harmony and the Dianchun Yi Pavilion demonstrates that the proposed framework reproduces wing corner geometry with sub-centimeter accuracy, providing a reusable technical approach for HBIM-based digital reconstruction of historic timber architecture.
Blockchain technology has emerged as a transformative solution for secure digital transactions, providing decentralized and tamper-resistant mechanisms for financial and asset trading. A wide range of blockchain-based trading mechanisms has been proposed across various applications, each employing distinct processes and different trading mechanism. However, assessing their comparative efficacy remains challenging due to this diversity, and there is a notable lack of comprehensive review studies addressing the issues. To bridge this gap, this paper presents a thorough review of blockchain-enabled trading mechanisms, focusing on critical trading parameters such as transparency, security, and efficiency. It also explores advanced models incorporating smart contracts, zero-collateral systems, and multi-level asset management strategies. Special attention is given to the trade-offs between decentralization and performance optimization within these systems. The review further examines the evolution of trading architectures, the role of consensus algorithms, and the application of cryptographic techniques in ensuring transaction integrity. Finally, the paper outlines current research trends, identifies open challenges, and suggests potential directions for future development, offering a comprehensive perspective on the strategic and technical advancements needed for secure and scalable blockchain-based trading platforms.
This paper focuses on a decision support system that leverages artificial intelligence to evaluate the editorial quality of textbooks distributed to Brazilian public schools through the Brazilian Textbook Program. The program is crucial for enhancing educational resources for students and educators throughout Brazil. Our study relies on a dataset comprising images presenting a subset of textbook image characteristics to tackle a multiclass problem, categorizing them into three classes: sharp, defocused-blurred, and motion-blurred. We conducted experiments using the Fourier and Haar transforms to evaluate their impact on the performance of our CNN classification model. By using Fourier and 10-fold cross-validation, the mean accuracy of our models varied from 75.44% to 80.25%, the mean precision varied from 78.38% to 83.87%, the mean recall varied from 76.05% to 80.37%, and the mean Area Under the Curve (AUC) varied from 90.27% to 92.14%. By using Haar, the mean accuracy of our models varied from 69.08% to 87.14%, the mean precision varied from 78.38% to 87.80%, the mean recall varied from 76.05% to 87.25%, and the mean AUC varied from 90.27% to 93.95%. These results indicate that employing CNNs with data preprocessed using the Haar transform method can achieve more consistent results for assessing textbook image quality.
