AIDC Microgrid Vulnerability Assessment Under Computing-Power Coordinated Attacks

Ze Yu, Hongwei Zhen, Chao Shen, Mingyang Sun

arXiv preprint, 2026.

• This work explores computing-power coordinated attacks against low-carbon AIDC microgrids, jointly assessing inverter control tampering and AI-induced demand manipulation under renewable and demand-response uncertainty.

Workload reshaping and waste heat valorization for token cost reduction in large model datacenters: A two stage forecasting and optimization framework

Junyan Shao, Chao Shen, Zihan Guo, Yujia Huang, Mingyang Sun

Applied Energy Under Review, 2026.

• The first work links token-level workload reshaping with waste heat valorization, opening a new optimization pathway for cost-efficient large-model datacenters.

OptArgus: A Multi-Agent System to Detect Hallucinations in LLM-based Optimization Modeling

Zhong Li, Zihan Guo, Xiaohan Lu, Juntao Wang, Jie Song, Chao Shen, Jiageng Wu, Mingyang Sun

NeurIPS 2026 Under Review, 2026.

• OptArgus is among the first systems to formalize optimization-modeling hallucination detection, using specialist agents to catch structural errors missed by objective-value checks.

Universal transient stability analysis: A pre-trained generative transformer-enabled power system dynamics prediction framework

Chao Shen, Ke Zuo, Mingyang Sun

Applied Energy, 2026. (Power/Energy Top, SCI Q1, IF 12.2)

• Uni-TSA presents a first universal transient-stability prediction framework that can generalize across operating conditions, fault scenarios, and even unseen power systems.

LLM-Guided Safe Reinforcement Learning for Energy System Topology Reconfiguration

Zongyan Zhang, Chao Shen, Xu Wan, Jie Song, Mingyang Sun

Applied Energy, 2026. (Power/Energy Top, SCI Q1, IF 12.2)

• This work pioneers an LLM-guided safe RL paradigm for topology reconfiguration, injecting domain reasoning into safety-critical switching decisions.

ProOPF: Benchmarking and Improving LLMs for Professional-Grade Power Systems Optimization Modeling

Chao Shen, Zihan Guo, Xu Wan, Zhen Yang, Yifan Zhang, Wen Huang, Jie Song, Zongyang Zhang, Mingyang Sun

International Conference on Machine Learning (ICML), 2026. (CCF-A)

• ProOPF introduces the first professional-grade OPF benchmark for LLMs, moving text-to-optimization evaluation from toy tasks to realistic power-system modeling.

LLM-DMD: Large Language Model-based Power System Dynamic Model Discovery

Chao Shen, Zihan Guo, Ke Zuo, Wen Huang, Mingyang Sun

IEEE Transactions on Industrial Informatics, 2026. (Automation/AI Top, SCI Q1, IF 9.8)

• LLM-DMD is an early attempt to use LLM agents for power-system dynamic model discovery, jointly searching differential equations and algebraic constraints.

Probabilistic Robustness Verified Data-Driven Transient Security-Constrained Optimal Power Flow

Ke Zuo, Chao Shen, Peng Cheng, Jie Song, Mingyang Sun

IEEE Transactions on Power Systems, 2025. (Power/Energy Top, SCI Q2, IF 8.7)

• This study advances data-driven TSCOPF by adding probabilistic robustness verification, making learned security constraints more reliable under uncertainty.

Physics-augmented auxiliary learning for power system transient stability assessment

Chao Shen, Ke Zuo, Mingyang Sun

IEEE Transactions on Industrial Informatics, 2025. (Automation/AI Top, SCI Q1, IF 9.8)

• PA-AL introduces physics-augmented auxiliary learning for TSA, using electrical velocity as a new auxiliary signal to improve both accuracy and physical consistency.

Physics-following neural network for online dynamic security assessment

Chao Shen, Ke Zuo, Mingyang Sun

IEEE Transactions on Power Systems, 2025. (Power/Energy Top, SCI Q2, IF 8.7)

• PFNN moves beyond generic physics-informed learning by explicitly following power-system dynamics for online security assessment.

Carbon-NeuGC: Neural Granger Causality Based Attribution Analysis of Power System Carbon Intensity

Chao Shen, Fengzhou Sun, Hao Chen, Yi Lin, Chuangxin Guo, Mingyang Sun

China Automation Congress (CAC), 2024.

• Carbon-NeuGC provides a neural Granger-causality framework for attributing power-system carbon intensity, enabling more interpretable carbon-flow analysis.