Electromagnetic transient (EMT) simulation is essential for transient stability analysis in renewable energy power systems, but its high computational cost limits large scale scenario screening, control tuning, and rapid post event assessment. This paper presents a surrogate-assisted EMT-based approach to enhance the efficiency of transient stability studies, where data-driven surrogate models are used to assist, rather than replace, EMT simulations. Three representative EMT-based tasks are investigated. For preevent analysis, voltage-observable surrogate models are used to approximate EMT-derived severity indicators and support efficient vulnerability ranking. For in-event analysis, surrogate-assisted ordinal optimization is employed to tune converter control parameters to improve fault ride-through (FRT) performance. For post-event analysis, stability-diagnosis models are developed using externally measurable voltage waveforms to detect instability and identify parameter interactions that influence it. Case studies on a representative renewable energy power system show that the proposed approach substantially reduces the number of EMT simulations while preserving key nonlinear transient characteristics relevant to stability assessment.
This paper proposes an approach for diagnosability analysis and enforcement of λ-free Petri nets. First, a verifier net is constructed based on the system model and its fault-free subnet. Then, the reachability graph of the verifier net is constructed, and a necessary and sufficient condition is established to determine whether the system is diagnosable. If the system is not diagnosable, a diagnosability enforcement strategy is further proposed. Specifically, we present a systematic procedure to construct a new labeling function that ensures the diagnosability of the system while minimizing cost. The construction follows a set of labeling rules, and an integer linear programming model is employed to obtain the optimal labeling scheme.
The physical implementation of embodied intelligence depends on high performance, energy-efficient, and reliable hardware systems that are deeply integrated with physical bodies. Gallium nitride (GaN) is a key material among the third generation of wide-bandgap semiconductors. It provides an ideal material platform for developing new multifunctional integrated devices thanks to its inherent advantages, including optoelectronic integration, high electron mobility, and a high breakdown electric field. GaN-based dual-mode devices integrate two or more dynamically switchable functional modes in a single physical structure. These devices offer a new way to overcome the hardware limitations in embodied intelligence, especially in sensing, computing, communication, and actuation. This review provides a systematic overview of recent progress in GaN-based dual-mode devices for supporting embodied intelligence. It explains the core operational mechanisms and application goals of these devices and analyzes the physical mechanisms that enable them to perform dual-mode functions. These mechanisms allow the devices to smoothly switch between different roles, such as fast detection and slow memory, high power output and high energy efficiency, and signal transmission and reception. In addition, this review highlights the potential of GaN dual-mode devices in practical applications of embodied intelligence and suggests future research directions for their use in large-scale systems.
Artificial intelligence (AI) tutoring systems have transformative potential for higher education, yet their integration faces critical challenges including but not limited to curriculum misalignment, epistemic unreliability, and fragmented learning experiences. This paper introduces the AI learning companion, a platform developed by the Department of Automation at Tsinghua University. The platform enforces course-specific knowledge isolation while explicitly modeling conceptual overlaps across courses via a curriculum knowledge graph. It integrates three key functional modules—multimodal resource ingestion, retrieval augmented intelligent Q&A with citation tracking, and automated quiz generation. It is powered by a layered architecture that leverages large language models through disciplined prompt engineering, context management, and evidence centered generation. Deployed across 36 courses and serving over 1300 students, our findings demonstrate that the platform enhances learning efficiency, fosters traceable and in-depth understanding, and can be integrated into the teaching-management loop for data-driven intervention.
The increasing complexity of modern power systems has created a strong demand for advanced data-driven tools. Generative artificial intelligence (GenAI) has been applied to a range of problems, including power flow analysis, stability analysis, fault diagnosis, and power system planning and operation. Recent studies have explored generative adversarial network (GAN)-, variational autoencoder (VAE)-, diffusion-, and large language model (LLM)-based approaches across these domains, but a systematic and unified review remains lacking. This survey fills that gap with a structured overview of GenAI architectures and their applications in power flow analysis, stability analysis, fault diagnosis, and planning and operation. Three principal findings emerge from this analysis. Tier 1 (established): In data-centric support tasks such as scenario generation, data augmentation, measurement reconstruction, and natural-language-to-model interfaces, GenAI provides consistent advantages over conventional methods. Tier 2 (conditional): In core analytical tasks such as classical transient stability assessment, AC power flow approximation, and component-level fault diagnosis, GenAI achieves competitive performance. Physical constraint enforcement and cross-system generalization, however, remain open problems that condition practical adoption. Tier 3 (not yet application-ready): Safety-critical autonomous real-time grid control and analysis of converter-dominated wideband dynamics remain out of reach for current GenAI methods. Scarce domain-specific data, the absence of formal feasibility guarantees, and strict latency requirements combine to form a fundamental barrier. These findings show what GenAI may or may not yet contribute to power systems and motivate the challenges and future directions outlined in the final section.
This paper explores the implementation of traveling wave protection (TWP) in microgrids through the integration of Internet of Things (IoT) technologies and a spiking recurrent neural network (SRNN). Microgrids present unique fault detection challenges, as conventional protection techniques can be hindered by reduced fault currents, bidirectional power flow, and communication latency. By leveraging high-frequency traveling wave signals, TWP offers rapid and precise fault localization. In parallel, IoT-enabled sensing provides real-time data acquisition and decentralized decision-making. The proposed SRNN further enhances fault classification and location accuracy by combining spiking neuron dynamics with recurrent memory. Hardware-in-the-loop (HIL) experiments on both simplified and complex microgrids demonstrate the method�s effectiveness in minimizing misclassification while maintaining low latency and reduced power consumption. This work extends our previous IoT-based TWP research by adopting a neuromorphic framework suitable for microgrid edge deployments, paving the way for more adaptive and robust protection solutions in modern distribution networks.
The intermittent nature of photovoltaic (PV) power poses significant challenges to grid operations. However, accurate forecasting is difficult since PV power depends on weather conditions that fluctuate within minutes. To address this, a novel multilayer weather classification-based regression model (MWCR) is developed to predict the PV output power for the coming 24 h in different datasets with 1-h and 5-min as time steps. The key idea is to classify weather conditions at each time step to capture the rapid PV power fluctuations. To address the absence of irradiance sensors, the power ratio of predicted power obtained by a long-short term memory (LSTM) model over the maximum power in a clear sky is used to reflect sky condition. Classification is based on power ratio, local irradiance, temperature, wind speed, and humidity. Then, for each class, a regression model with the same inputs is developed. To improve the accuracy of the model, an LSTM model predicting power is integrated as an additional input to the regression model. Prediction results of three datasets at 1-h and 5-min time steps demonstrate superior accuracy for both models compared to an LSTM model alone in all testing datasets, especially on a day with high PV power fluctuations.
This paper investigates the multi-objective scheduling problem of intelligent unmanned operations and analyzes the interdependent constraints among the various links of the operation workflow. To address the coupled challenges of job sequencing and resource allocation inherent in unmanned operation scenarios, an integrated scheduling model based on a two-layer particle swarm optimization framework is proposed. A mixed-integer programming formulation is adopted to rigorously characterize the structural constraints and logical dependencies within the scheduling process. Building upon this model, an enhanced two-layer particle swarm optimization algorithm with fragment-based particle encoding is introduced to expand the feasible search space. Moreover, a dynamic inertia weight adjustment mechanism and an adaptive mutation operator are incorporated to strengthen the algorithm's global exploration capability while accelerating convergence. Simulation experiments verify that the proposed model and algorithm effectively optimize the scheduling of unmanned operations under complex operational constraints, significantly improving system-level support performance. These results demonstrate that the method provides a robust and efficient solution for multi-objective intelligent scheduling tasks in highly constrained unmanned operation environments.
The control barrier function (CBF) method is becoming a popular tool that transforms nonlinear constrained optimal control problems into a sequence of quadratic programs (QPs). In this tutorial paper, we show how to employ machine learning techniques to ensure the feasibility of these QPs, which is a challenging problem, especially for high relative degree constraints where high order CBFs (HOCBFs) are employed. We present two complementary learning approaches: (i) parameter learning for regular unsafe sets; (ii) sampling learning for irregular unsafe sets, where “regularity” of an unsafe set is formally defined in terms of the dependence of QP feasibility on initial system conditions. The first approach compensates for the myopic nature of the QP based approach by parameterizing the HOCBFs and using machine learning techniques to select parameters that maximize a feasibility robustness metric related to system performance. This feasibility robustness metric measures the extent to which QP feasibility is maintained in the presence of time-varying and unknown unsafe sets. The sampling learning approach addresses “irregular” unsafe sets in which the problem feasibility heavily depends on the initial conditions. This approach learns a new feasibility constraint that guarantees the QP feasibility, and it is then enforced by another HOCBF added to the QPs. The accuracy of the learned feasibility constraint can be recursively improved by the proposed recurrent training algorithm. We demonstrate the advantages of the proposed learning approaches to constrained optimal control problems with specific focus on a robot control problem and on autonomous driving in an unknown environment.
Asymmetry naturally exists in real-life applications, such as directed graphs. Different from classical kernel methods requiring Mercer kernels with symmetry and semi-positive definiteness, Kernel Singular Value Decomposition (KSVD) is proposed recently as a powerful tool that extends SVD to nonlinear feature spaces and is capable of directly tackling asymmetric kernels. As generally in kernel methods, KSVD also suffers from inefficiency with large-size data, but can be significantly sped up by the asymmetric Nyström method, facilitating the scalability. KSVD is handled under the framework of kernel methods with well derived primal-dual representations, benefiting practitioners from exploring asymmetry in generic feature learning. The insight and methodology in KSVD pave the way for further explorations of asymmetric kernel methods in machine learning and beyond. There are different ways to derive KSVD, where important common properties and fundamental differences w.r.t. classical Mercer kernels are revealed towards in-depth understandings and promising future developments of asymmetric kernel methods. This keynote presents the first systematic introduction to the modeling, optimization, and practicality of the timely proposed KSVD. In this keynote, the formulations of KSVD with both kernels and covariance operators are first reviewed with detailed derivations, covering rigorous discussions on the current applications and promising future outlooks. Then, the connections to existing kernel methods are discussed comprehensively, aiming to present in-depth understandings on KSVD and its potentials.
The transition to Industry 4.0 is promoting personalized production, challenging traditional design methods reliant on manual expertise and lengthy iterations. Although generative AI shows promise for automation, its industrial use is limited by difficulties in capturing and applying complex rule-based design specifications. To address this, we present "AutoFrit", a comprehensive parametric automobile frit dataset containing design pairs from major manufacturers. Our evaluation shows that models trained on AutoFrit drastically reduce design iteration time from months to minutes while ensuring adherence to industrial standards, effectively addressing the scalability challenges in modern manufacturing.
Charging optimization is a key challenge to the implementation of quantum batteries, particularly under inhomogeneity and partial observability. This paper employs reinforcement learning to optimize piecewise-constant charging policies for an inhomogeneous Dicke battery. We systematically compare policies across four observability regimes, from full-state access to experimentally accessible observables (energies of individual two-level systems (TLSs), first-order averages, and second-order correlations). Simulation results demonstrate that full observability yields near-optimal ergotropy with low variability, while under partial observability, access to only single-TLS energies or energies plus first-order averages lags behind the fully observed baseline. However, augmenting partial observations with second-order correlations recovers most of the gap, reaching 94
Over the past decade, deep learning models have exhibited considerable advancements, reaching or even exceeding human-level performance in a range of visual perception tasks. This remarkable progress has sparked interest in applying deep networks to real-world applications, such as autonomous vehicles, mobile devices, robotics, and edge computing. However, the challenge remains that state-of-the-art models usually demand significant computational resources, leading to impractical power consumption, latency, or carbon emissions in real-world scenarios. This trade-off between effectiveness and efficiency has catalyzed the emergence of a new research focus: computationally efficient deep learning, which strives to achieve satisfactory performance while minimizing the computational cost during inference. This review offers an extensive analysis of this rapidly evolving field by examining four key areas: 1) the development of static or dynamic light-weighted backbone models for the efficient extraction of discriminative deep representations; 2) the specialized network architectures or algorithms tailored for specific computer vision tasks; 3) the techniques employed for compressing deep learning models; and 4) the strategies for deploying efficient deep networks on hardware platforms. Additionally, we provide a systematic discussion on the critical challenges faced in this domain, such as network architecture design, training schemes, practical efficiency, and more realistic model compression approaches, as well as potential future research directions.
Enabled by rapidly developing quantum technologies, it is possible to network quantum systems at a much larger scale in the near future. To deal with non-Markovian dynamics that is prevalent in solid-state devices, we propose a general transfer function based framework for modeling linear quantum networks, in which signal flow graphs are applied to characterize the network topology by flow of quantum signals. We define a noncommutative ring D and use its elements to construct Hamiltonians, transformations and transfer functions for both active and passive systems. The signal flow graph obtained for direct and indirect coherent quantum feedback systems clearly show the feedback loop via bidirectional signal flows. Importantly, the transfer function from input to output field is derived for non-Markovian quantum systems with colored inputs, from which the Markovian input-output relation can be easily obtained as a limiting case. Moreover, the transfer function possesses a symmetry structure that is analogous to the well-know scattering transformation in Schrödinger picture. Finally, we show that these transfer functions can be integrated to build complex feedback networks via interconnections, serial products and feedback, which may include either direct or indirect coherent feedback loops, and transfer functions between quantum signal nodes can be calculated by the Riegle's matrix gain rule. The theory paves the way for modeling, analyzing and synthesizing non-Markovian linear quantum feedback networks in the frequency-domain.
Instrumentation and Measurement (I&M) is a field which is constantly developing due to the emergence of new technologies. In recent years, with the rapid development in computer hardware and computational power, Artificial Intelligence (AI) has demonstrated remarkable successes in data analytics, thus offering new paradigm for the design and applications of new instruments and measurement systems. The applications of AI to I&M have made measurements of some quantities in industry possible or more cost-effective. This paper presents a review of recent AI based methods applied in different aspects of instrumentation systems for monitoring complex industrial processes with a particular focus on multiphase flow metering, combustion monitoring as well as carbon dioxide flow measurement under carbon capture and storage conditions. This review also explores how AI is playing an important role in expanding knowledge across the whole spectrum of I&M. Trends and future developments of AI methods in the field of I&M are also discussed.
The operational safety of large-scale dynamic systems in complex and constantly changing environments is paramount. Therefore, safety assessment techniques are essential to help avoid accidents and ensure safe operation. This paper first reviews different definitions and perspectives on safety assessment for dynamic systems. In particular, we formulate the online safety assessment (OSA) problem to attract more attention from researchers. Then, the current state of research on system safety assessment is presented, including commonly used methods and application objects. The safety assessment methods in the literature are categorized into offline and online approaches. However, offline approaches are primarily knowledge-based and cannot be applied in real-time. Meanwhile, research on data-driven OSA methods is still in its early stages and faces various challenges. These challenges are summarized as human-in-the-loop, domain knowledge integration, complex data characteristics, and open environments. Finally, this paper provides insight into promising directions for the further development of OSA.
The relationship between genetic variation and human phenotypes is crucial for developing effective treatments and personalized medicine. However, our understanding of the regulatory mechanisms by which variants influence human traits and diseases is far from complete. Context-specific regulatory network is a typical tool that provides detailed understanding of gene regulation in specific biological contexts, allowing us to identify key regulators and pathways that are important for a particular phenotype. In this review, we summarize the large international biobanks and reference omics data that provide diverse datasets for the genotype-phenotype analysis and the construction of context-specific regulatory networks, and discuss the importance of context-specific regulatory networks in explaining the underlying causal mechanism between genotypes and phenotypes. We emphasize the significance of quantitative trait locus (QTL) studies in explaining the correlation between genotypes and omics features, and present various computational approaches for the construction of context-specific regulatory networks. With continued advancements in biobanking, genomics, and computational biology, the context-specific regulatory networks may serve as an increasingly powerful tool for modeling the causal mechanisms that underlie the relationship between genotypes and phenotypes.
