Parameter-Efficient Domain Adaptation of Physics-Informed Self-Attention based GNNs for AC Power Flow Prediction
Pattern Recognition Lab, Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany(模式识别实验室,埃尔兰根-纽伦堡大学,埃尔兰根,德国);Institute of Electrical Energy Systems, Friedrich-Alexander-Universität Erlangen-Nürnberg, Germany(电气能源系统研究所,埃尔兰根-纽伦堡大学,德国)
Redwanul Karim, Changhun Kim, Timon Conrad, Nora Gourmelon, Julian Oelhaf, David Riebesel, Tomás Arias-Vergara, Andreas Maier, Johann Jäger, Siming Bayer
在中压(MV)电网训练的模型部署到高压(HV)网络时,准确的交流电力流(AC-PF)预测在领域转移下至关重要。现有的物理信息图神经网络(GNN)求解器通常依赖全微调进行跨领域转移,导致高再训练成本,并且对目标领域适应与源领域保留之间的稳定性-可塑性权衡控制有限。我们研究了物理信息自注意力基于GNN的参数高效领域适应,通过物理基础损失鼓励基尔霍夫一致行为,同时限制适应为低秩更新。具体而言,我们应用低秩适应(LoRA)到注意力投影,并选择性地解冻预测头以调节适应能力。这种设计在电压领域转移下实现了可控的效率-精度权衡。在多个电网拓扑结构上,所提出的LoRA+PHead适应方法在目标领域RMSE差距为$2.6 imes 10^{-4}$的情况下恢复了接近全微调的精度,同时将可训练参数数量减少了$85.46\%$。物理基础残差与全微调相当;然而,相对于全微调,LoRA+PHead在领域转移下将中压源保留减少了4.7个百分点(17.9% vs. 22.6%),同时仍实现了参数高效且物理一致的AC-PF估计。
英文摘要:
Accurate AC power flow (AC-PF) prediction under domain shift is critical when models trained on medium-voltage (MV) grids are deployed on high-voltage (HV) networks. Existing physics-informed graph neural network (GNN) solvers typically rely on full fine-tuning for cross-regime transfer, incurring high retraining cost and offering limited control over the stability-plasticity trade-off between target-domain adaptation and source-domain retention. We study parameter-efficient domain adaptation for physics-informed self-attention-based GNNs, encouraging Kirchhoff-consistent behavior via a physics-based loss while restricting adaptation to low-rank updates. Specifically, we apply low-rank adaptation (LoRA) to attention projections with selective unfreezing of the prediction head to regulate adaptation capacity. This design yields a controllable efficiency-accuracy trade-off for physics-constrained inverse estimation under voltage-regime shift. Across multiple grid topologies, the proposed LoRA+PHead adaptation recovers near-full fine-tuning accuracy with a target-domain RMSE gap of $2.6 \times 10^{-4}$ while reducing the number of trainable parameters by $85.46\%$. The physics-based residual remains comparable to full fine-tuning; however, relative to Full FT, LoRA+PHead reduces MV source retention by 4.7 percentage points (17.9% vs. 22.6%) under domain shift, while still enabling parameter-efficient and physically consistent AC-PF estimation.
Transfer Learning in Physics-Informed Neural Networks: Full Fine-Tuning, Lightweight Fine-Tuning, and Low-Rank Adaptation
Tsinghua University(清华大学);Bauhaus-Universität Weimar(魏玛包豪斯大学);Leibniz University Hannover(汉诺威莱布尼茨大学);CNPC Engineering Technology RD Company Limited(中国石油集团工程技术研究院有限公司);Queensland University of Technology(昆士兰科技大学)
Towards Foundation Model for Chemical Reactor Modeling: Meta-Learning with Physics-Informed Adaptation
Department of Chemical and Biomolecular Engineering, National University of Singapore(化学与生物分子工程系,新加坡国立大学);Artificial Intelligence Institute, National University of Singapore(人工智能研究所,新加坡国立大学)
QuanTA: Efficient High-Rank Fine-Tuning of LLMs with Quantum-Informed Tensor Adaptation
NSF AI Institute for Artificial Intelligence and Fundamental Interactions(NSF人工智能与基础交互AI研究所);Massachusetts Institute of Technology(麻省理工学院);Harvard University(哈佛大学)
$φ$-Adapt: A Physics-Informed Adaptation Learning Approach to 2D Quantum Material Discovery
Department of Electrical Engineering and Computer Science, University of Arkansas(电气工程与计算机科学系,阿肯色大学);Department of Physics, Montana State University(物理系,蒙塔纳州立大学);Department of Physics, University of Arkansas(物理系,阿肯色大学)
Physics-informed and Unsupervised Riemannian Domain Adaptation for Machine Learning on Heterogeneous EEG Datasets
University Paris-Saclay(巴黎萨克雷大学);Inria(法国国家信息与自动化研究所);CEA(法国原子能和替代能源委员会);TAU Inria(特拉维夫大学-法国国家信息与自动化研究所联合机构);LISN-CNRS(法国国家科学研究中心LISN实验室);Roche Pharma Research and Early Development(罗氏制药研发部);Roche Innovation Center Basel(罗氏创新中心巴塞尔分部);F. Hoffmann–La Roche Ltd(罗氏控股有限公司)
GIPSO: Geometrically Informed Propagation for Online Adaptation in 3D LiDAR Segmentation
University of Trento(特伦托大学);Télécom-Paris, Intitute Polytechnique de Paris(巴黎理工学院电信巴黎);Sapienza University of Rome(罗马第一大学);NVIDIA AI Technology Center(英伟达人工智能技术中心);Fondazione Bruno Kessler(布鲁诺·凯斯勒基金会)
Reducing Oversmoothing through Informed Weight Initialization in Graph Neural Networks
School of Electrical Engineering and Computer Science, National Technical University of Athens(雅典国立技术大学电气与计算机工程学院);National Center for Scientific Research “Demokritos”(德谟克利特国家科学研究中心)
SA-GDA: Spectral Augmentation for Graph Domain Adaptation
Beijing Institute of Technology(北京理工大学);Michigan State University(密歇根州立大学);California State Polytechnic University, Pomona(加州州立大学,庞纳);Mohamed bin Zayed University of Artificial Intelligence(穆罕默德·本·扎耶德人工智能大学)
Rank and Align: Towards Effective Source-free Graph Domain Adaptation
School of Computer Science, National Key Laboratory for Multimedia Information Processing(多媒体信息处理国家重点实验室计算机学院);University of California, Los Angeles(加州大学洛杉矶分校);University of International Business and Economics(对外经济贸易大学);Anker Innovations(安克创新);Peking University(北京大学);Peking University-Anker Embodied AI Lab(北京大学-安克具身智能实验室)
针对无源图领域自适应问题,提出Rank and Align (RNA)方法,通过谱序列化进行鲁棒语义学习,并利用对抗性边采样提取域不变子图以减少域差异,实验证明其有效。
Smoothness Really Matters: A Simple Yet Effective Approach for Unsupervised Graph Domain Adaptation
Institute of Artificial Intelligence, Beihang University(北京航空航天大学人工智能研究院);Institute of Computing Technology, Chinese Academy of Sciences(中国科学院计算技术研究所);Zhongguancun Laboratory(中关村实验室)
Transformers for molecular property prediction: Domain adaptation efficiently improves performance
Spoken Language Systems, Saarland Informatics Campus, Saarland University(spoken language systems,萨尔兰信息学校园,萨尔兰大学);Medical Faculty, Saarland University(医学学院,萨尔兰大学);Center for Bioinformatics, Saarland University(生物信息学中心,萨尔兰大学);Data Driven Drug Design, Center for Bioinformatics, Saarland University(数据驱动药物设计,生物信息学中心,萨尔兰大学)
Advancing Heatwave Forecasting via Distribution Informed-Graph Neural Networks (DI-GNNs): Integrating Extreme Value Theory with GNNs
Institute for Resources, Environment and Sustainability, University of British Columbia(不列颠哥伦比亚大学资源、环境与可持续性研究所);Environment and Climate Change Canada(加拿大环境与气候变化部);Department of Earth, Ocean and Atmospheric Sciences, University of British Columbia(不列颠哥伦比亚大学地球、海洋与大气科学系);Department of Forestry, University of British Columbia(不列颠哥伦比亚大学林业系);Department of Occupational Science and Occupational Therapy, University of British Columbia(不列颠哥伦比亚大学职业科学与职业治疗系);Department of Space Science, Institute of Space Technology(空间技术研究院空间科学系);Peaceful Society, Science and Innovation Foundation(和平学会科学与创新基金会)
RBF-MGN:Solving spatiotemporal PDEs with Physics-informed Graph Neural Network
College of Aerospace Science and Engineering, National University of Defense Technology(国防科技大学航天科学与工程学院);National Innovation Institute of Defense Technology, Chinese Academy of Military Science(中国人民解放军军事科学院国防科技创新研究院)
Label-informed Graph Structure Learning for Node Classification
Center for Research on Intelligent Perception and Computing, Institute of Automation, Chinese Academy of Sciences(中国科学院自动化研究所智能感知与计算研究中心);School of Artificial Intelligence, University of Chinese Academy of Sciences(中国科学院大学人工智能学院)
Architecture--Optimization Co-Design for Physics-Informed Neural Networks via Layer-wise Coordinate Adaptation and Gradient Conflict Resolution
College of Applied Mathematics(应用数学学院);Chengdu University of Information Technology(成都信息科技大學);College of Mathematics(数学学院);Sichuan University(四川大学);School of Sciences(科学学院)