
Industrial Internet of Things (IIoT)-enabled vehicular communication has become an important infrastructure for smart transportation, intelligent logistics, and industrial mobility applications. In such scenarios, vehicles, roadside units, sensors, and infrastructure nodes are interconnected to support reliable information exchange, while the limited energy supply of distributed vehicular devices remains a critical bottleneck. To enhance the reliability and energy sustainability of IIoT-enabled vehicular networks, this article investigates a reconfigurable intelligent surface (RIS)-assisted decode-and-forwardcooperative simultaneous wireless information and power transfer (SWIPT) system. Considering cochannel interference at both the relay and destination nodes, the outage performance of the proposed system is analyzed under time switching relay (TSR) and power splitting relay (PSR) protocols. Semianalytical outage expressions are evaluated by Gauss–Laguerre quadrature and verified by Monte Carlo simulations. Furthermore, the impacts of the number of RIS reflecting elements, outage threshold, Nakagami-m fading parameter, time switching factor, power splitting factor, and energy conversion efficiency are examined. Numerical results show that RIS deployment improves reliability under the considered mean-interference benchmark, whereas the relative performance of TSR and PSR depends on the allocation and channel parameters.
Accurate reading of integrated circuit (IC) markings is essential for product traceability and quality analysis in industrial visual inspection. In printed circuit board assembly images captured with a large field of view, substantial character-scale variation, highly regular structured backgrounds, and degraded marking details jointly hinder accurate detection and recognition. We propose DSR-ICMark, a unified framework for IC marking detection, restoration, and recognition. The detail-sensitive adaptive detector builds multiresolution representations and selects effective scales based on local content, preserving fine details while suppressing background interference. The character-balanced map loss combines region-level importance with prediction confidence to adaptively calibrate spatial supervision and focus optimization on sparse foregrounds and challenging regions. A lightweight character restoration network reconstructs damaged strokes and local details.The recognition-guided fidelity optimization strategy uses downstream task supervision to guide restoration, reducing the effects of image degradation on feature discriminability. Extensive benchmarks show that DSR-ICMark matches or outperforms representative detection, recognition, and spotting models, with fewer parameters and faster inference. Industrial deployment tests show that DSR-ICMark achieves the lowest end-to-end latency, supporting its potential for practical IC marking inspection.
With the increasing penetration of renewable energy sources, power systems are characterized by lower inertia and greater uncertainty, thereby increasing the risk of frequency violations. To address this challenge, this article proposes an adversarial max–min robust scheduling framework that co-optimizes dynamic frequency security and battery energy storage system (BESS) degradation in renewable-rich power systems. First, a two-stage analytical frequency-response model is developed to capture both unsaturated and reserve-limited primary frequency response dynamics and to derive explicit constraints on the rate of change of frequency (RoCoF) and frequency nadir. In addition, a rainflow-based BESS cycle-aging model is established to quantify the lifetime impact of repeated frequency-support operation. To solve the resulting nonlinear max–min problem efficiently, deep neural network surrogate models are embedded through a big-M reformulation and combined with a combinatorial cutting algorithm to obtain a tractable mixed-integer linear programmingformulation. Case studies demonstrate that the proposed framework maintains the RoCoF within 0.15 Hz/s and the frequency nadir at or above 49.4 Hz throughout the operating horizon. Moreover, explicitly accounting for BESS degradation reduces the monthly degradation cost from USD 65 000 to USD 36 000, corresponding to a 44.6% reduction, while maintaining a comparable overall system operating cost.
Although numerous advanced active suspension control methods have been developed, few studies address the inherent conflict between ride comfort and road holding. Most existing approaches focus on single-objective optimization without considering these two conflicting performance indices simultaneously and achieving a compromise between them. This article presents a Pareto-based output-feedback optimal control method for active suspension systems, where the Pareto front can be obtained to achieve a tradeoff between ride comfort and road holding. Specifically, two independent value functions associated with ride comfort and road holding are designed. Then, an output algebraic Riccati equation (OARE) is derived to avoid using full system states and observer techniques. To obtain the optimal solution of the OARE, an adaptive learning algorithm is designed to ensure the convergence to the optimal solution. Considering the conflicting nature of ride comfort and road holding, the Pareto dominance principle is employed, by which the Pareto-based output-feedback optimal control gain can be calculated based on the derived optimal solution of the OARE. Hence, the Pareto front can be characterized to provide a series of feasible solutions for different active suspension control requirements. Theoretical analysis is conducted to prove the existence of the Pareto-based output-feedback optimal control gains. Simulation and experimental results validate the effectiveness of the proposed method.
Accurate real-time estimation of power deficits following disturbances is critical for maintaining grid stability in low-inertia power systems with high renewable energy penetration. This article presents a transformer-enhanced hybrid model-data-driven framework that synergistically combines physical interpretability with adaptive learning for real-time power deficit estimation. The approach consists of two co-designed components: first, an improved model-driven module employing sliding-window Gauss–Newton frequency fitting to extract center-of-inertia rate-of-change-of-frequency for generator-side estimation; and second, a novel Transformer with multihead frequency attention mechanism that captures complex nonlinear load-side dynamics through multiscale feature extraction. The hybrid framework incorporates scenario-specific adaptive correction mechanisms that learn optimal fusion weights, preserving interpretability while enhancing accuracy. Validated on the Chinese Society for Electrical Engineering—Frequency Stability benchmark and a provincial receiving-end power grid, both with high renewable-energy penetration, the proposed framework outperforms representative deep-learning baselines [long short-term memory (LSTM), long short-term memory–convolutional neural network (LSTM-CNN), and LSTM-CNN with an attention mechanism (lstm-CNN-Attention)] in mean relative error, maximum relative error, and root-mean-squared error. The framework delivers reliable real-time performance for practical grid-control applications.