Kaiyuan Pang

Voltage and reactive power control in active distribution networks is a safety-critical and highly dynamic problem, where traditional optimization methods often struggle to balance efficiency and robustness under complex operating conditions. Recently, large language models (LLMs) have shown promise in sequential decision-making tasks, but their direct application to power system control remains limited by the lack of physical grounding and safety guarantees. In this paper, we propose SafeVolt, a closed-loop LLM-based framework that integrates multi-candidate action generation, simulator-in-the-loop evaluation, and a fine-tuned expert judge for safety-aware decision making. In addition, a high-level rule distillation mechanism that converts successful control experiences into reusable operational axioms is introduced to enable iterative self-improvement. Experiments on a standard distribution network scenario demonstrate that the proposed method outperforms representative baselines, achieving substantial improvements in average reward, voltage violation rate, reactive power loss, and system stability. In particular, voltage violations and extreme events are substantially reduced, indicating enhanced operational safety. These results suggest that combining LLM reasoning with physical simulation and structured feedback provides a promising direction for reliable and adaptive power system control.

In the power industry, how to efficiently and reliably query relevant documents has always posed a challenge for electrical professionals. Unreliable or inefficient query results can lead to significant inefficiencies and introduce unpredictable errors. Hence, a reliable and efficient knowledge querying system is critical. In practice, the effectiveness of Graph-based Retrieval-Augmented Generation (RAG) systems lies in providing expressive representation of entities and graph structures and this makes it stand out as a widely-used approach for document retrieval. However, typical GraphRAG frameworks encounter challenges such as semantic dilution and topological drift caused by generic technical terminology and granular graph noise especially in professional documents like regulations, etc, which is one of the mostly used type of document in electric industry. Thus, we propose KG-Anchored RAG, a framework that shifts the retrieval paradigm from community-based summarization to precision-guided anchoring. During knowledge construction, our framework employs a topological skeleton refinement and constructs a Knowledge Attachment Matrix using latent topic modeling and one-hot feature injection. During inference, non-linear sharpening and PageRank-based structural resonance are utilized to locate high-density knowledge cells. Evaluation on professional documents in the power industry reveals that our method outperforms localized search baselines in terms of context precision, generative faithfulness, and ranking quality. The proposed framework demonstrates a superior ability to prioritize evidentiary clauses and reduce information redundancy without relying on computationally expensive external re-rankers. Experimental results indicate that KG-Anchored RAG effectively mitigates speculative hallucinations, establishes a reliable architectural paradigm for retrieval-augmented generation in high-stakes, safety-critical vertical industries.

Traditional data-driven residential short-term load forecasting methods are faced with two major challenges, i.e., preserving data privacy and addressing missing data. Regarding privacy-preserving, a new two-stage pre-training personalized federated learning framework is proposed to train customized forecasting models with high forecasting accuracy for all households without violating their data privacy. As for the missing data problem, an advanced semi-supervised graph diffusion network model is proposed to address missing “features” and “labels” problems, where “features” and “labels” refer to the required input data of the forecasting model and actual load data of the forecasting day, respectively. Mathematically, a graph feature diffusion mechanism-based feature input and imputation module is developed to mitigate the deficiency imposed by missing “features”. Next, a graph convolutional network-based semi-supervised feature augmented and update module is established to mitigate the negative effects of missing “labels”. An actual power system in China is employed to demonstrate the proposed method, and simulation results indicate that high forecasting accuracy and strong robustness under missing data scenarios are achieved.

Frequency response rate (FRR) constraints are imposed during synchronous generator-dominated distribution system restoration (DSR) to prevent unallowable frequency dip caused by load pickup. However, with the ever-increasing penetration of diversified inverter-based resources (IBRs), the FRR is evolving as the synthetic FRR (SFRR), rendering their constraints more intractable to be incorporated in DSR. This letter proposes a novel method for losslessly imposed fine-grained SFRR constraints on IBR-rich DSR optimization models, for the first time. Simulation results verify that it can accurately ensure the secure synthetic frequency response in each formed microgrid through concise linearized constraints, without order reduction of frequency transfer functions or piecewise approximations of frequency nadir.

This article presents a dynamic restoration strategy for active distribution networks (ADNs) by coordinating repair crew dispatch and frequency-constrained cold load pickup. To incorporate the stochastic repair time, the repair crew dispatch is formulated as “event-driven” with the implementation of model predictive control (MPC). The stochastic repair time is estimated, convexified, and updated dynamically with each MPC execution. The finish of a repair task triggers the subsequent cold load pickup model, where the frequency dynamics are computed and linearly constrained with the help of a uniform frequency response model for low-inertia systems. Next, a co-optimization framework of the two models is developed to coordinate the repair crew dispatch and cold load pickup under a unified time scale. Numerical results on a modified IEEE 33-node test feeder and a real-world 136-node distribution system have verified the effectiveness of the proposed model.

The post-disruption microgrid (MG) formation and the subsequent scheduling are resilience-enhancing measures for active distribution networks (ADNs) against disastrous events. This article proposes an integrated MG formation and scheduling solution, considering stochastic loads and distributed generators (DGs). Specifically, a first-stage MG formation model and a second-stage MG scheduling model are proposed to flexibly identify MG boundaries and restore as many critical loads as possible. The MG formation problem is formulated over multiple time steps to address time-varying generation outputs of intermittent DGs. Moreover, in the MG scheduling problem, the stochasticity of load demands and DG outputs is modeled as joint chance constraints (JCCs) to describe the possibility of security violations in the overall system. An efficient approximation method is next developed to convexify the JCCs into solvable second-order cones. The interaction between the MG formation and scheduling by the model predictive control implementation achieves a responsive solution for enhancing the emergency support capability of ADNs by incorporating power forecasts and real-time scheduling. The performance of the proposed method is verified in a modified real-world 136-node test feeder under catastrophic failures.

This paper proposes a post-disaster cyber-physical interdependent restoration scheduling (CPIRS) framework for active distribution networks (ADN) where the simultaneous damages on cyber and physical networks are considered. The ad hoc wireless device-to-device (D2D) communication is leveraged, for the first time, to establish cyber networks instantly after the disaster to support ADN restoration. The repair and operation crew dispatching, the remote-controlled network reconfiguration and the system operation with DERs can be effectively coordinated under the cyber-physical interactions. The uncertain outputs of renewable energy resources (RESs) are represented by budget-constrained polyhedral uncertainty sets. Through implementing linearization techniques on disjunctive expressions, a monolithic mixed-integer linear programming (MILP) based two-stage robust optimization model is formulated and subsequently solved by a customized column-and-constraint generation (C&CG) algorithm. Numerical results on the IEEE 123-node distribution system demonstrate the effectiveness and superiorities of the proposed CPIRS method for ADN.

This paper proposes a two-stage method for the robust design of resilient active distribution networks (ADNs) against high-impact and low-probability (HILP) events. The line hardening and the deployment of remote-controlled switches (RCSs) are considered as two powerful measures for resilience enhancement. Especially, the hardening of tie lines and the deployment of bilateral tie switches are emphasized as part of the resilient design. A novel progressive detection mechanism (PDM) is devised to estimate the potential propagation of outages and identify surviving nodes outside of the minimum outage area after intentional islanding. The proposed PDM method considers potential RCS locations among regular lines and tie lines on outage to calculate the global optimal design scheme. The two-stage robust design model is formulated as a mixed-integer linear programming (MILP). The nested column-and-constraint generation (nested C&CG) algorithm is customized to solve the proposed model. Numerical results on a modified IEEE 33-node distribution system demonstrate the effectiveness and the superiority of the proposed resilience enhancement method.

Forming microgrids (MG) in distribution systems (DSs) is an operational measure proposed to enhance power system flexibility. The computational performance of the applied optimization techniques varies greatly according to the constraints ensuring network radiality. Formulations of radiality constraints differ from each other by model complexity and the search space of the relaxed linear programming (LP) problem. In this paper, widely used formulations of radiality constraints are first revisited and flexibly enhanced to address the complex operating environment of modern DSs. To incorporate various radiality constraints and provide an optimization basis for comparison, a flexible MG formation problem is proposed with the ability to automatically allocate distributed generators (DGs) into multiple MGs. Combined formulations with parent-child (PC) constraints are then developed to shrink the LP feasible region. Moreover, to accelerate the branching process by applying the branch-and-cut (B&C) algorithm, an efficient formulation of radiality constraints is proposed by adding extra constraints that exclude fundamental loops into the single-commodity flow formulation. To further accelerate the branching process, a dynamic constraint generation algorithm is developed without substantially increasing the complexity of the LP problem. DSs of various complexities and sizes are employed to verify the computational efficiency of the proposed formulations. The paper concludes by suggesting the most efficient formulations of radiality constraints according to the scale of the targeted DS.