Search Results (709)

The rapid expansion of sub-6 GHz 5G and Internet of Things (IoT) networks demands highly miniaturized Multiple-Input Multiple-Output (MIMO) antennas. However, balancing extreme physical compactness with rigorous inter-port isolation introduces severe computational bottlenecks for conventional optimization algorithms. To overcome these multidimensional challenges, this paper proposes a novel Deep Surrogate Active Learning framework for the autonomous design and empirical validation of an ultra-compact dual-band MIMO antenna. By using a surrogate-assisted closed-loop strategy to reduce reliance on repeated full-wave evaluations, the methodology combined a custom-penalized Deep Neural Network with dynamic boundary reduction. After training the initial surrogate model with 440 valid full-wave responses obtained from the offline design-of-experiments (DOE) stage, the best CST-validated candidate was identified at the 83rd active learning cycle. The optimized nested-loop geometry, incorporating a partial defected ground structure (DGS), occupies an extremely confined footprint of only 1634 mm² on a Rogers RO4350B substrate (Rogers Corporation, Chandler, AZ, USA). The selected geometry provided simulated −10 dB impedance bands of 3.35–3.88 GHz and 4.34–5.05 GHz, while the complete two-port model maintained inter-port isolation better than 13.8 dB and 14.9 dB across the lower and upper target passbands, respectively. Measurements of the fabricated prototype showed the intended dual-band behavior, a maximum measured gain of 4.54 dBi, and total radiation efficiencies of approximately 51–63% across both ports at the evaluated frequencies. The simulated Envelope Correlation Coefficient (ECC) remained below 0.035 across the target passbands, supporting the suitability of the compact geometry for the investigated sub-6 GHz MIMO bands. Full article

In low-voltage distribution networks, load switching, induction motor start-up, photovoltaic output variations, and short-circuit faults may produce highly overlapping electrical characteristics, which can lead to maloperation or failure to operate in conventional protection. To address this problem, this paper proposes an adaptive protection method integrating physically guided and cost-sensitive learning. First, an incremental topology-constraint deviation and a voltage-current trajectory curvature are constructed based on the fault-superimposed network constraint and the variation characteristics of system equivalent impedance, enabling the discrimination of short-circuit faults from non-fault transient disturbances. Then, a cost-sensitive physically guided extreme gradient boosting (XGBoost) model is developed, in which a fault-current-increment-based weight is introduced into the objective function to enhance the learning capability for weak-fault samples. Furthermore, a temporal-consistency-based protection operation logic is designed using sliding-window confirmation and majority voting to suppress isolated abnormal predictions. Simulation and RTDS-based real-time validation results on a 0.4-kV low-voltage distribution network with distributed photovoltaic generation show that the proposed method improves weak-fault detection sensitivity and reduces maloperation under complex source–load disturbances. The method relies only on local measurements and has potential for deployment in low-voltage intelligent protection terminals. Full article

The increasing penetration of renewable energy sources has increased the need for advanced control strategies capable of maintaining stability under low-inertia, converter-dominated operating conditions. In grid-connected wind energy conversion systems (WECSs), constant power loads (CPLs) exhibit negative incremental impedance characteristics that can amplify DC-link oscillations and complicate the coordination between the electrical and mechanical subsystems. The main contribution of this work is a Soft Actor–Critic (SAC) reinforcement learning algorithm that tunes the outer proportional-integral gains of the machine-side DC-voltage-squared control loop together with the active damping gain, allowing online adaptation of the controller according to the operating condition and disturbance level, thereby improving energy system sustainability. The proposed control framework includes a two-mass shaft model, virtual inertia control, and DC-link load uncertainty in the form of both resistive loads and CPLs. The system is modeled and evaluated using MATLAB/Simulink, and its performance is compared with that of a conventional fixed-gain controller under AC load disturbances and wind speed variations. It has been found that for a 25% load disturbance, the maximum DC-link voltage deviation is reduced by 1.2% under resistive loading and 6.5% under CPL operation. For a 1 m/s reduction in wind speed, the corresponding reductions are 0.8% and 0.9%, respectively. The proposed controller also provides smoother output power and improved damping of the rotor speed and system frequency responses. Full article

Digital twin (DT) technology has emerged as a cornerstone of Industry 4.0, facilitating real-time synchronization between physical assets and virtual models to drive operational excellence. Unlike prior surveys that address singular industrial domains, presenting qualifications in general terms without paradigm-to-task mapping, this review uniquely synthesizes DT architectural maturation across Technology Readiness Levels (TRLs) 1 through 9, quantitative performance outcomes from 39 documented industrial implementations spanning 10 sectors, and an explicit algorithmic taxonomy mapping distinct AI paradigms to specific functional DT requirements. By synthesizing empirical data across the aerospace, automotive, and manufacturing sectors, this study evaluates the quantitative impact of DT implementation, highlighting significant gains in predictive maintenance, production efficiency, and design cycle reduction. This research further examines the synergistic role of Machine Learning (ML) paradigms integrated within DT systems, specifically, physics-informed neural networks (PINNS), generative adversarial networks (GANs), deep transfer learning, reinforcement learning, and federated learning, in enhancing diagnostic accuracy and enabling autonomous decision-making. Despite these advancements, this review identifies critical barriers in data interoperability, cybersecurity, and workforce expertise that impede widespread adoption. This paper concludes by outlining future research directions, emphasizing the necessity for standardized data protocols and secure, distributed DT ecosystems to unlock the full potential of cyber-physical integration in a data-driven industrial landscape. Full article

The systemic shift in the energy consumption structure from high-carbon fossil fuels to low-carbon clean energy constitutes a critical pathway toward global climate governance and carbon neutrality. However, this sustainable transition is consistently impeded by deep-seated institutional frictions and structural barriers, such as governance fragmentation and carbon lock-in effects embedded in traditional industrial organization. Whether digital government development can overcome these barriers by nurturing resilient business ecosystems and thereby promote a systemic low-carbon energy transition remains an urgent question within sustainable development research. To address this issue, this study integrates digital government development, regional e-commerce ecosystem competitiveness, and the low-carbon transition of the energy consumption structure into a unified analytical and sustainable governance framework. Using panel data from 30 Chinese provinces from 2012 to 2022, we exploit the institutional reform of provincial big data administrations as a quasi-natural experiment to identify the impacts of digital government. Regional e-commerce ecosystem competitiveness is comprehensively evaluated across four sustainable dimensions: ecological innovation capacity, market connectivity, ecological global integration, and inclusive infrastructure. Methodologically, we employ a spatial difference-in-differences model to capture geographic interdependencies alongside a double machine learning framework to handle high-dimensional confounding and nonlinear disturbances. The empirical findings reveal that both digital government development and regional e-commerce ecosystem competitiveness significantly drive the low-carbon transition of the energy consumption structure. The institutional effect of digital government exhibits strong regional embeddedness with localized impacts, whereas e-commerce ecosystem competitiveness generates positive spatial spillovers that accelerate energy optimization in neighboring regions. Crucially, regional e-commerce ecosystem competitiveness serves as a significant partial mediator, constructing a reliable transmission channel from institutional design to market-based decarbonization. Further pathway analysis indicates that market connectivity and inclusive infrastructure function as the primary transmission channels, effectively mitigating transportation energy intensity and bridging the digital-green divide, while the mediating contribution of ecological innovation capacity is relatively constrained due to cross-organizational coordination thresholds. This study clarifies the interactive mechanism between public digital governance and market ecosystem competitiveness in advancing environmental sustainability, thereby offering fresh theoretical insights and actionable policy implications for emerging market economies striving for economic growth and decarbonization. Full article

This applied systems-integration study addresses the “Impedance Mismatch” inherent in multi-energy microgrids—namely, the conflict where disparate physical domains traditionally force the use of isolated or highly rigid Supervisory Control and Data Acquisition (SCADA) data silos. We present a scalable, open-source edge-to-cloud telemetry pipeline orchestrated entirely through a custom MING+React stack (Mosquitto, InfluxDB, Node-RED, Grafana). The core architectural contribution is a category-specific Canonical Data Model (CDM) that functions as a translation layer, effectively decoupling sensor hardware from ingestion logic. Coupled with an automated Null-Pruning middleware loop and Metadata Separation, this architecture maintains high input purity prior to time-series persistence. To validate the system, the middleware was subjected to a high-fidelity stochastic simulation utilizing a Strict Corridor Algorithm to mimic physical inertia across 12 distinct energy vectors. Simulation-based validation demonstrates that under simulated conditions, the pipeline maintains data integrity for valid telemetry packets, achieves high accuracy in pruning malformed data, and operates with low latency under concurrency. These findings demonstrate the feasibility of this applied architecture as a resilient, cross-domain research environment for modern Digital Learning Factories. Full article

Force control is crucial for robotic contact-rich tasks such as assembly, grinding, and polishing, directly affecting task accuracy, interaction stability, and safety. Yet in unstructured environments, environmental uncertainty, contact oscillations, and friction disturbances make high-performance contact control and reinforcement learning policy optimization difficult. To address this issue, this paper proposes a deep reinforcement learning-based variable-damping impedance control method that integrates parameterized Hindsight Experience Replay (TO-HER) and Importance Sampling (IS). Within the impedance control framework, a residual parameterized policy enables online damping adjustment, improving dynamic adaptability across contact phases. To overcome the training instability of conventional HER in contact-intensive tasks, a parameterized goal relabeling mechanism is introduced to improve relabeled sample quality and sample efficiency. In addition, an importance sampling scheme based on density ratio estimation mitigates the distribution mismatch between relabeled and real samples, enhancing training quality and stability. Experimental results show that the proposed method outperforms baseline methods in convergence, final return, and training stability, while achieving higher tracking accuracy and better dynamic response in typical contact scenarios, demonstrating strong effectiveness and robustness for robotic contact control in unstructured environments. Full article

Electrochemical Impedance Spectroscopy (EIS) is a non-destructive technique for characterizing the battery behavior for estimating the state of health (SOH). EIS provides frequency-domain information on key parameters, including solid electrolyte interphase (SEI) resistance, charge-transfer (CT) resistance, and ohmic resistance, which are sensitive to battery degradation mechanisms. In an EIS test, a sinusoidal excitation signal is applied to the battery, and the corresponding voltage response is analyzed to extract the impedance spectrum. The reliability of SOH estimation therefore depends critically on the accurate and repeatable extraction of impedance features. This paper investigates the variability in impedance spectra arising from the state of charge (SOC), temperature, rest time, and repeated measurements under nominally identical conditions. This variability is identified as drift and represents previously underexplored variations in the impedance spectrum. To quantify these variations, this work proposes a normalized resistance-based index that captures changes in the impedance spectrum using estimated equivalent circuit model (ECM) parameters. The proposed index is applicable across battery chemistries, sizes, and operating conditions. It is evaluated using published datasets spanning different chemistries, SOC levels, and temperatures, as well as laboratory data collected from repeated EIS experiments. The results show that even at fixed SOC and temperature, repeated measurements can produce measurable bias and variance in ECM parameters. These findings highlight the importance of accounting for drift in EIS analysis and motivate uncertainty-aware battery diagnostics for practical SOH monitoring systems. Full article

Reliable state of health (SOH) and remaining useful life (RUL) estimation is essential for lithium-ion battery diagnostics, prognosis, and management across cell, module, and pack levels. Yet the reported performance metrics often remain tied to controlled cell datasets, with batteries degrading due to chemistry shifts, protocol variation, temperature changes, inconsistent and insufficient measurements, and pack-level heterogeneity. This critical review investigates what evidence is required before an SOH/RUL estimator can be considered trustworthy, transferable, and suitable for battery management system deployment. Based on a de-duplicated classified set of 176 scientific works and a supplementary evidence audit workbook, this review synthesizes model-based, machine learning, deep learning, transfer learning, physics-informed, impedance-based, thermographic, relaxation-based, and digital twin approaches through observability, robustness, uncertainty calibration, transferability, and deployment feasibility. A compact mathematical framework formalizes the health inference, domain shift, cross-fidelity degradation, calibrated uncertainty, and BMS-facing validation criteria. The analysis argues that deployment-ready battery health intelligence should be evaluated as an evidence system rather than as a point prediction task. The proposed multi-fidelity validation framework links synthetic cells, controlled aging, module (pack) testing, fleet shadow operation, and closed-loop safety-governed deployment using acceptance criteria, based on worst-domain error, calibration data, warning risk, and computational feasibility. Full article

Accurate monitoring of oil palm plantations is important for balancing economic development with environmental conservation in Southeast Asia. However, existing plantation maps often suffer from low spatial resolution and a lack of recent temporal coverage, impeding effective surveillance of rapid land-use changes. In this study, we propose a deep learning framework to generate 10 m resolution oil palm plantation maps for Indonesia and Malaysia from 2020 to 2024, utilizing Sentinel-2 imagery without requiring new manual annotations. To address the resolution mismatch between coarse 100 m historical labels and 10 m imagery, we employ a U-Net architecture optimized with Determinant-based Mutual Information (DMI). This approach effectively mitigates the influence of label noise. We validated our method against 2058 manually verified points, achieving overall accuracies of 70.64%, 63.53%, and 60.06% for the years 2020, 2022, and 2024, respectively. The gradual decline in accuracy with time is consistent with a growing temporal mismatch between the 2016 historical reference labels and the later prediction years. At the regional scale, the mapped oil palm area suggests a peak in 2022 followed by a lower mapped extent in 2024. Land cover transition analysis further indicates exchanges with cropland and flooded vegetation, which should be interpreted together with the reported accuracy and uncertainty. Given the moderate per-year accuracies and the temporal mismatch between the 2016 supervision and the later prediction years, these results should be interpreted as regional-scale indicators rather than pixel-level change maps. The generated maps can support regional monitoring, sustainability assessment, and prioritization of areas for further validation. Full article

Noninvasive blood glucose monitoring using impedance spectroscopy faces challenges due to high-dimensional, redundant data and complex nonlinear relationships with glucose concentration. To address these challenges, LASSO regression was applied for feature selection, followed by ensemble learning models incorporating multiple base learners. Among them, a Stacking ensemble framework—combining Support Vector Regression (SVR), Random Forest (RF), and LightGBM—achieved the best overall regression accuracy (MAE = 0.9662, RMSE = 1.3040, $R^2$ = 0.9414). Compared to the best individual base learner (MLP), the S1 ensemble improved overall MAE by 9.7% and RMSE by 7.6%. However, in the clinically critical hyperglycemic range, the individual MLP substantially outperformed S1 (RMSE: 0.7570 vs. 1.1716), indicating that S1’s superiority reflects a global statistical average rather than uniform improvement across all glycemic subranges. Therefore, while the proposed framework demonstrates the efficacy of combining feature selection with ensemble learning, the MLP model may be preferable for clinical applications prioritizing hyperglycemic risk identification. Full article

The integration of artificial intelligence (AI) and machine learning (ML) is fundamentally transforming pharmaceutical sciences, shifting drug formulation and nanocarrier design from traditional empirical approaches toward predictive, data-driven methodologies. By enabling the analysis of large, complex datasets, AI technologies are accelerating decision-making, improving formulation efficiency, and supporting the development of more effective therapeutic systems. Despite these advances, the successful clinical translation of advanced nanomedicines, including polymeric nanoparticles and mRNA–lipid nanoparticle platforms, remains limited by challenges such as biological barriers, highly sensitive formulation parameters, scalability issues, and the limited interpretability of many computational models. This review provides a comprehensive overview of AI applications throughout the pharmaceutical development lifecycle. It explores how classical machine learning algorithms and deep learning architectures optimize conventional dosage forms, enhance formulation development, and enable the rational design of targeted nanocarriers. Particular emphasis is placed on predicting critical quality attributes, encapsulation efficiency, physicochemical properties, drug-release behavior, therapeutic efficacy, and early-stage nanotoxicity. Furthermore, we critically assess the regulatory considerations, manufacturing constraints, data quality issues, and tumor microenvironment heterogeneity that continue to impede bench-to-clinic translation. Ultimately, overcoming these challenges requires moving beyond isolated algorithmic optimization toward an integrated framework that combines computational intelligence, robust experimental validation, and continuous clinical feedback. Such a synergistic approach is expected to drive the next generation of precision nanomedicine and facilitate the safe and effective translation of AI-enabled pharmaceutical innovations into clinical practice. Full article

Understanding how climate risk impedes agricultural technology innovation, along with its underlying mechanisms, is critical for promoting high-quality agricultural development. Utilizing a panel dataset of 223 prefecture-level cities in China from 2005 to 2023, this study employs a fixed effects panel OLS estimator with city and year fixed effects to systematically examine the impact of climate risk on agricultural technology innovation. The findings indicate that climate risk significantly inhibits agricultural technology innovation, with a baseline coefficient of −1.462 that is statistically significant at the 1% level. These findings remain robust after addressing endogeneity concerns and applying Double Machine Learning (DML). Heterogeneity analysis reveals that this inhibitory effect is weaker in plain regions, major grain-producing areas, and regions with advanced digital economies. Mechanistically, climate risk suppresses agricultural technology innovation by reducing capital supply and dampening market demand. Furthermore, logistics capacity, knowledge accumulation, and human capital mobility can mitigate this negative impact. Additional analysis indicates that climate risk also negatively affects innovation quality, collaborative synergy, and green transition, albeit to a lesser extent than its impact on overall innovation levels. Within green innovation, climate risk exerts a stronger inhibitory effect on ecological-enhancing innovation than on pollution-reducing innovation. Full article

Electromagnetic wave absorbing materials are important for electromagnetic protection, radar stealth, wireless communication, and advanced electronic systems. However, traditional design methods mainly rely on repeated experiments and full-wave simulations, which are time-consuming and inefficient when dealing with complex compositions, microstructures, and multilayer structures. Machine learning provides a new route to accelerate the design of high-performance absorbers by learning the relationship among material composition, structure, electromagnetic parameters, and absorption performance. This review summarizes recent progress in machine-learning-empowered electromagnetic wave absorbing materials. First, the basic physical principles of electromagnetic wave absorption are introduced, including reflection loss, impedance matching, attenuation, and physical limits such as the Rozanov and Snoek limits. Then, typical machine learning models are discussed, including classical machine learning, deep learning, generative models, physics-informed models, large language models, and artificial-intelligence (AI) Agents. Their applications are further summarized from forward property prediction, high-throughput screening, inverse design, electromagnetic parameter decoupling, physics-informed modeling, explainability, multi-objective optimization, and data augmentation. Finally, the main challenges and future directions are discussed, including data standardization, physics-guided learning, foundation models, autonomous laboratories, and engineering-scale validation. This review shows that machine learning is changing absorber research from experience-driven trial-and-error to data-driven and knowledge-driven design, and provides a useful reference for developing next-generation electromagnetic wave absorbing materials. Full article

This paper proposes a safe and effective human–robot physical interaction control framework for exoskeleton robots that enhances system compliance and safety while enabling the robot to adapt to human motion. The framework is designed around two primary objectives: first, a model-free adaptive control method is employed for reference trajectory estimation to achieve real-time estimation of human motion intention; second, the Forgetting Factor Recursive Least Squares (FFRLS) method is utilized for online estimation and the learning of human impedance parameters, considering their time-varying nature. In addition, a model-free adaptive trajectory tracking control strategy is proposed to optimize control performance during human–robot physical interaction. Simulation results demonstrate that the proposed control framework outperforms conventional methods significantly in terms of safety and compliance. Full article