Lizhou Jiang

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.

The rapid proliferation of Large Language Models (LLMs) is increasing electricity demand from data centers, creating new challenges for power-demand forecasting and grid planning. A key difficulty is that architecture- and deployment-related information that affects inference load is often private to LLM providers. This paper proposes a two-stage, mechanism-assisted forecasting framework under information asymmetry. In the first stage, a stylized incentive mechanism elicits verifiable reduced-form demand parameters from LLM providers at a chosen reporting precision. In the second stage, the elicited parameters are incorporated into forecasting models as architecture- and deployment-informed features. Using calibrated synthetic scenarios constructed from public data-center energy reports, open LLM-inference energy benchmarks, and secondary public estimates, we find that incorporating elicited parameters reduces the mean squared error (MSE) of the ResNet forecasting backbone by 65.1% relative to an architecture-agnostic ResNet baseline. Similar improvements are observed for a gradient-boosting model, indicating that the main empirical value comes from procuring informative provider-side demand features rather than from a specific neural architecture. The results should be interpreted as a proof-of-concept demonstration rather than a full operational model of LLM serving or power-system dispatch.

With the increasing integration of high-penetration renewable energy and emerging loads, power system operation simulation faces two major challenges, namely strong uncertainty and significant heterogeneity in the output characteristics of multiple generator types. Traditional mathematical programming methods struggle to effectively handle uncertainty while meeting real-time computational requirements. Existing deep learning approaches fail to decouple the heterogeneous output characteristics of different generator types, which limits their ability to achieve coordinated operation. To address these issues, this paper proposes a task-decoupled and multi-task synergistic LLM-MoE method for power system operation simulation. First, a feature encoder based on Residual-Gated Linear Units is constructed to perform deep filtering and efficient representation of multi-source heterogeneous data. Second, a pre-trained large language model is employed as a temporal feature extractor to enhance temporal modeling capability and cross-scenario generalization. Finally, a customized gating-controlled mixture-of-experts decoder is developed. It dynamically coordinates task-specific and shared experts, which enables unified modeling of task decoupling, cross-task information sharing, and system physical constraints. Simulation results based on a provincial-level power grid in China demonstrate that the proposed method achieves high-accuracy and high-efficiency operation simulation while ensuring physical consistency.

Power documentation serves as the core guideline for the safe operation of power systems, and its precise retrieval is crucial for ensuring grid stability and safety. In this context, Retrieval-Augmented Generation (RAG) frameworks emerge as an effective technique by combining LLMs with natural language understanding capabilities and a retrieval-based model with traceability. However, existing Retrieval-Augmented Generation (RAG) frameworks face several main challenges for power-system documents: semantic drift caused by non-standardized industry terminology, increased semantic noise due to fixed-window segmentation, and knowledge conflicts in the multi-source retrieval context. To address these challenges, we propose a multi-path adaptive fusion retrieval framework based on small language models (SLMs). To map queries to standard terminology, our framework first constructs a common terminology repository and section-structure-aware index for the power industry while fully preserving the physical hierarchical logic from related documents. Subsequently, the SLM in our framework assigns prior weights based on query features and retrieved context, which contributes to adaptive fusion of retrieval paths through confidence assessment and consistency verification. With the help of the fusion process, our method effectively filters retrieval noise and resolves knowledge conflicts. Experimental results on real-world power-document datasets covering dispatch, energy storage and emergency response show that our framework achieves an average recall of 91%, outperforming DENSE and BM25 by 21% and 28% respectively. Compared with other methods, it yields the optimal BERTScore F1 (0.7798) and Rouge-1/2/L F1 (0.2430, 0.1588, 0.2098) and achieves the best results in the RAGAS framework evaluation, which significantly enhances the rigor and reliability of the question-answering system in the power engineering domain.

This paper addresses the challenges of scarce high-risk scenario samples in power grid operation and the difficulty of traditional methods to balance overall distribution rationality with specific feature requirements. A power grid scenario generation method based on a prior knowledge embedded conditional generative adversarial network is proposed. The method encodes operational risk features such as node overvoltage and line power flow overload as conditional variables. A feature-aware loss function is constructed to embed physical constraints into the training objective of generative adversarial networks. This approach achieves organic integration of data-driven learning and knowledge-driven guidance. Case studies demonstrate that the proposed method significantly improves the generation ratio of risk scenarios at designated locations and types while maintaining the reasonableness of overall data distribution. This provides data support with both physical interpretability and computational efficiency for power grid security analysis, risk assessment, and intelligent dispatching.