GENCO、LUMINA、GridSFM 三款电网基础模型泛化能力测试:超出训练条件后表现接近需要微调的元模型
Generalization Testing of Grid Foundation Models for Power Flow and Optimal Power Flow
电力系统做潮流和最优潮流的三款基础模型被统一放上台面测试,结论挺扎心:线路变长后 LUMINA 和微调版 GENCO 干不过 DC 基线,GridSFM 在陌生电网上直接失效。做电力 AI 或考虑采购这类模型的人应该看看这套三轴测试方法。
arXiv 论文(编号 2611.19194)提出一个三轴泛化测试框架,用于评估电网基础模型在潮流计算与最优潮流任务上的能力边界:改变训练电网的导纳矩阵、改变负载水平、以及测试完全未见过的电网。研究团队将 GENCO、LUMINA 和 GridSFM 三款模型与 DC 基线对比测试。
实验结果显示了几项具体规律:在电网中新增一条线路对所有三个模型的损害大于移除一条线路;当线路变长时,LUMINA 和微调后的 GENCO 最终表现不如 DC 基线;三个模型在超出训练负载范围后都出现退化,且此时负载-电压曲线变陡。在未见过的电网上,LUMINA 即使在训练负载范围内也急剧退化,GridSFM 则在与训练电网距离最远的那张电网上直接失效。
论文据此给出的结论是:电网基础模型在训练条件之外的表现更像元模型,需要测试时的微调数据和算力才能维持精度。这对电力行业从业者意味着,现有 FM 尚不能作为跨电网通用求解器直接部署,在导纳结构变化、负载波动等实际工况下仍需依赖传统 DC 求解或额外微调。
Generalization Testing of Grid Foundation Models for Power Flow and Optimal Power Flow
There is growing interest in foundation models (FMs) for power grids, particularly for power flow and optimal power flow. Their usefulness depends on generalization accuracy, which existing works test on their own benchmarks. We propose a three-axis framework to test their generalization limits: changed admittance matrices of training grids, load levels, and unseen grids. We test GENCO, LUMINA and GridSFM against the DC baseline, alongside their reported limitations, and discuss theoretical reasons for their failures. Experiments reveal that adding a line degrades all FMs more than removing one, and longer lines eventually make LUMINA and fine-tuned GENCO worse than DC. All three degrade beyond their training load range, where the loading-voltage curve steepens. On unseen grids, LUMINA degrades sharply even inside its training range, and GridSFM fails on the grid farthest from it. Thus, beyond training conditions, FMs behave more like meta-models requiring test-time fine-tuning data and compute.
本条情报归属 AITOP「论文」核心追踪栏目。在跨区域产业观察坐标中,该动态直接关联「AI 开发者工具链、Coding Agent 与工程化实践」研究主线,反映出当前技术演进与商业落地的最新风向。涉及核心实体与技术议题包括:GENCO、LUMINA、GridSFM、电网基础模型,读者可通过下方标签进入对应专题页查阅历史脉络与关联对比。
信源核验说明:本文原始线索采编自「arXiv cs.LG」(@Ayushi Jolotia, Parikshit Pareek),经 AITOP 多维评分管线结构化提炼。