Active-control coils on Keda Torus eXperiment (KTX) are used to suppress error fields and mitigate MHD instabilities, thereby extending discharge duration and improving plasma confinement quality. Achieving effective active MHD control imposes stringent requirements on the coil power supplies: wide-bandwidth and high-precision current regulation, deterministic low-latency response, and tightly synchronized operation across 136 independently driven coils. Specifically, the supplies must deliver up to ±200 A with fast slew rates and bandwidths up to several kilohertz, while ensuring sub-100 μs control latency, programmable waveforms, and inter-channel synchronization for real-time feedback. These demands make the power supply architecture a key enabling technology and motivate this work. This paper presents the design and simulation of the KTX active-control coil power supply. The system adopts a modular AC–DC–AC topology with energy storage: grid-fed rectifiers charge DC-link capacitor banks, each H-bridge IGBT converter (20 kHz) independently drives one coil, and an EMC filter shapes the output current. Matlab/Simulink R2025b simulations under DC, sinusoidal, and arbitrary current references demonstrate rapid tracking up to the target bandwidth with ±0.5 A ripple at 200 A and limited DC-link voltage droop (≤10%) from an 800 V, 50 mF storage bank. The results verify the feasibility of the proposed scheme and provide a solid basis for real-time multi-coil active MHD control on KTX while reducing instantaneous grid loading through energy storage.

Massive Internet of Things (IoT) deployments face critical spectrum crowding and energy scarcity challenges. Energy harvesting (EH) symbiotic radio (SR), where secondary devices share spectrum and harvest energy from non-orthogonal multiple access (NOMA)-based primary systems, offers a sustainable solution. We consider long-term throughput maximization in an EHSR network with a nonlinear EH model. To solve this non-convex problem, we designed a two-layered optimization algorithm combining convex optimization with a deep reinforcement learning (DRL) framework. The derived optimal power, time allocation factor, and the time-varying environment state are fed into the proposed long short-term memory (LSTM) attention mechanism combined Deep Deterministic Policy Gradient, named the LAMDDPG algorithm to achieve the optimal long-term throughput. Simulation results demonstrate that by equipping the Actor with LSTM to capture temporal state and enhancing the Critic with channel-wise attention mechanism, namely Squeeze-and-Excitation Block, for precise Q-evaluation, the LAMDDPG algorithm achieves a faster convergence rate and optimal long-term throughput compared to the baseline algorithms. Moreover, we find the optimal number of PDs to maintain efficient network performance under NLPM, which is highly significant for guiding practical EHSR applications.

Electroencephalogram (EEG)-based emotion recognition has emerged as a key enabler for semantic communication systems in next-generation networks (5G-Advanced/6G), where the goal is to transmit task-relevant semantic information rather than raw signals. However, domain adaptation approaches for EEG emotion recognition typically assume closed-set label spaces and fail when unseen emotional classes arise, leading to negative transfer and degraded semantic fidelity. To address this challenge, we propose a Coarse-to-Fine Open-set Domain Adaptation (C2FDA) framework, which aligns with the semantic communication paradigm by extracting and transmitting only the emotion-related semantics necessary for task performance. C2FDA integrates a cognition-inspired spatio-temporal graph encoder with a coarse-to-fine sample separation pipeline and instance-weighted adversarial alignment. The framework distinguishes between known and unknown emotional states in the target domain, ensuring that only semantically relevant information is communicated, while novel states are flagged as unknown. Experiments on SEED, SEED-IV, and SEED-V datasets demonstrate that C2FDA achieves superior open-set adaptation performance, with average accuracies of 41.5% (SEED → SEED-IV), 42.6% (SEED → SEED-V), and 48.9% (SEED-IV → SEED-V), significantly outperforming state-of-the-art baselines. These results confirm that C2FDA provides a semantic communication-driven solution for robust EEG-based emotion recognition in 6G-oriented human–machine interaction scenarios.

Large-scale renewable power supply system design for remote hydrogen production is a challenging task due to the 100% power electronics sending-end subsystem. The proper grid-forming strategy for a sending-end system to achieve large-scale remote hydrogen production still remains a research gap. This study first designs two grid-forming strategies for the concerned renewable power supply system, with one being based on virtual synchronous generator (VSG) and another one being based on V/f control. Then, the impedance analysis is carried out for ensuring the small-signal stable operation of the sending-end system including wind power plant and PV plant. Numerical simulation results implemented on PSCAD verify that the VSG-based grid-forming strategy configured on the sending-end modular multilevel converter (MMC) station of the MMC-based high-voltage direct-current (HVDC) link has a larger transient stability margin. Hence, the MMC-HVDC-based grid-forming strategy is a better choice for the power supply of large-scale remote hydrogen production. The enhanced stability margin ensures more robust operation under disturbances, which is critical for maintaining continuous power supply to large-scale electrolyzers.

Two new active disturbance rejection control (ADRC) structures for nonlinear systems are introduced: a locally linearized variant and a fully nonlinear formulation. Both approaches incorporate model knowledge to enhance performance but differ in how nonlinear dynamics are integrated into the control and observer design. The first proposed structure employs a state-dependent local approximation of the nonlinear model to generate dynamic controller and observer gains, aiming to balance robustness and accuracy. In contrast, the second one directly embeds the full nonlinear dynamics into both the control law and extended state observer, tightly coupling performance to model fidelity. The proposed methods were experimentally validated on a magnetic levitation system, known for its strong nonlinearity, and compared with a classical linear ADRC (LADRC). Furthermore, stability analysis of the methods was conducted using Lyapunov theory. Results show that the linearized structure consistently improves regulation performance over LADRC and, in most cases, achieves similar results to nonlinear ADRC with lower computational effort. However, the performance of the nonlinear approach may degrade under modeling inaccuracies and limited observer bandwidth. This study highlights that the way model information is integrated–rather than its level of detail–has a decisive impact on control quality. Finally, practical design guidelines are provided to assist in selecting an appropriate ADRC structure for nonlinear applications where robustness, computational efficiency, and limited model knowledge must be balanced.

In large-scale IoT environments, two major challenges—limited edge storage resources and complex task dependencies—make efficient management of service placement and task offloading particularly difficult. Existing approaches often optimize these two aspects independently while overlooking their tight interrelationship, resulting in poor performance in dynamic settings. To address this co-optimization challenge, we propose a Hierarchical Deep Q-Network (HDQN) framework that simultaneously manages service placement and task offloading in task-dependent MEC systems. HDQN divides the decision process into two levels: a meta-controller for long-term service placement and resource planning, and a subcontroller that makes real-time task offloading decisions based on the latest system state. This two-layer structure enables the framework to efficiently adapt to changing conditions while meeting both dependency and resource constraints. Evaluation across diverse experimental conditions—including varying numbers of users, MEC servers, communication rates, and service types—demonstrates that our proposed HDQN framework achieves a significant enhancement in task latency optimization compared to mainstream advanced algorithms like DDPG and DQN, underscoring its superior performance.

The modular multilevel converter (MMC) has become a pivotal technology in high-voltage direct current (HVDC) transmission systems due to its modularity, superior harmonic performance, and enhanced controllability. However, conventional control strategies, including model predictive control (MPC) and sorting-based voltage balancing methods, often suffer from high computational complexity, limited real-time performance, and inadequate handling of transient events. To address these challenges, this paper proposes a novel Multi-Output Neural Network-based hybrid control strategy that integrates a multi-output neural network (MONN) with an optimized reduced-switching-frequency (RSF) sorting algorithm. The MONN directly outputs precise submodule switching signals, eliminating the need for traditional sorting processes and significantly reducing switching losses. Meanwhile, the RSF algorithm further minimizes unnecessary switching operations while maintaining voltage balance. Furthermore, to enhance the accuracy of predicted switching stage, we extend the MONN for submodule activation count prediction (ACP) and employ a novel Cardinality-Constrained Post-Inference Projection (CCPIP) to further align the predicted switching stages and activation count. Simulation results under dynamic load conditions demonstrate that the proposed method achieves a 76.1% reduction in switching frequency compared to conventional bubble sort, with high switch prediction accuracy (up to 92.01%). This approach offers a computationally efficient, scalable, and adaptive solution for real-time MMC control, enhancing both dynamic response and steady-state stability.

Most control strategies for magnetic bearings are typically formulated upon the linearized suspension force model, and the nonlinear characteristics are neglected or regarded as the disturbance and variation in parameters of the control system. The controllers based on linearized suspension force model struggle to achieve fast response under disturbance. Therefore, a nonlinear mathematic model that simultaneously represents the main nonlinearity of suspension force and facilitates the design of high-performance controller is necessary to establish. In this study, a new mathematical model of suspension force with varying current stiffness is developed, and a specific controller was designed based on this model. Firstly, the nonlinear mathematical model of six-pole radial–axial hybrid magnetic bearing (RAHMB) is established. Secondly, the characteristics of the current stiffness varying with rotor displacement are analyzed and the expression between current stiffness and rotor displacement is fitted. Then, the linearized model is built via Taylor expansion of the nonlinear model. Subsequently, the varying current stiffness model is constructed by substituting the fitted expression of varying current stiffness into linearized model. Finally, an adaptive nonsingular terminal sliding mode controller (ANTSMC) is designed based on the proposed varying current stiffness model. The simulation and experiment results have shown that the ANTSMC based on varying current stiffness model reduces chattering more than 64% and reduces convergence time more than 70% to the NTSMC based on the linearized model.

In the original publication [...]

Industrial visual inspection demands high-precision anomaly detection amid scarce annotations and unseen defects. This paper introduces a zero-shot framework leveraging multimodal feature fusion and stabilized attention pooling. CLIP’s global semantic embeddings are hierarchically aligned with DINOv2’s multi-scale structural features via a Dual-Modality Attention (DMA) mechanism, enabling effective cross-modal knowledge transfer for capturing macro- and micro-anomalies. A Stabilized Attention-based Pooling (SAP) module adaptively aggregates discriminative representations using self-generated anomaly heatmaps, enhancing localization accuracy and mitigating feature dilution. Trained solely in auxiliary datasets with multi-task segmentation and contrastive losses, the approach requires no target-domain samples. Extensive evaluation across seven benchmarks (MVTec AD, VisA, BTAD, MPDD, KSDD, DAGM, DTD-Synthetic) demonstrates state-of-the-art performance, achieving 93.4% image-level AUROC, 94.3% AP, 96.9% pixel-level AUROC, and 92.4% AUPRO on average. Ablation studies confirm the efficacy of DMA and SAP, while qualitative results highlight superior boundary precision and noise suppression. The framework offers a scalable, annotation-efficient solution for real-world industrial anomaly detection.