| 引用本文: | 王振国,李 特,郑文哲,等.物理知识引导下覆冰迭代自迁移预测[J].电力系统保护与控制,2025,53(20):95-105.[点击复制] |
| WANG Zhenguo,LI Te,ZHENG Wenzhe,et al.Physics knowledge guided iterative self-transfer learning for icing prediction[J].Power System Protection and Control,2025,53(20):95-105[点击复制] |
|
| 摘要: |
| 高影响覆冰事件对电力系统的稳定运行构成威胁。然而，在小样本场景下，传统的物理知识引导和数据模型驱动的预测效果欠佳。因此，将物理信息神经网络(physics-informed neural network, PINN)与自迁移学习(self-transfer learning, STL)融合，提出一种物理知识引导下覆冰迭代自迁移(physics knowledge guided iterative self-transfer learning, PKG-ISTL)模型。首先，将数据划分为源域及目标域，构建包含空间、特征与时间维度的三维张量，实现滑动窗口覆冰预测。其次，将PINN与STL融合搭建模型。在源域分支，训练具备物理知识的PINN指导模型。在自迁移分支，应用多核最大均值差异进行域自适应处理。在目标域分支，运用知识蒸馏将专家模型中的物理知识自迁移至受训模型。最后，利用广西省某区段多个输电线路的历史覆冰数据进行算例仿真，并通过可解释性分析，揭示气象、力学、线路及覆冰因素等对线路覆冰的影响程度。结果表明，相比传统数据模型驱动，PKG-ISTL模型预测精度提升47.69%，验证了其在小样本场景的有效性。 |
| 关键词: 覆冰预测 物理信息神经网络 自迁移学习 知识蒸馏 三维张量 |
| DOI：10.19783/j.cnki.pspc.241682 |
| 投稿时间：2024-12-17修订日期：2025-03-24 |
| 基金项目:国家自然科学基金重点项目资助(U22B20106);浙江省自然科学基金项目资助(LZJMY25D050006)；国网浙江省电力有限公司科技项目资助(B311DS24001A) |
|
| Physics knowledge guided iterative self-transfer learning for icing prediction |
| WANG Zhenguo1,LI Te1,ZHENG Wenzhe1,WANG Yan2,HOU Hui2,LIN Xiangning3 |
| (1. Research Institute of State Grid Zhejiang Electric Power Company, Hangzhou 310014, China; 2. School of Automation,
Wuhan University of Technology, Wuhan 430070, China; 3. School of Electrical and Electronic Engineering,
Huazhong University of Science and Technology, Wuhan 430074, China) |
| Abstract: |
| High impact icing events pose a threat to the stable operation of power systems. However, under small-sample scenarios, traditional physics-knowledge guidance and data-driven prediction methods often have poor performance. Therefore, a physics knowledge guided iterative self-transfer learning (PKG-ISTL) model is proposed, integrating physics informed neural network (PINN) with self-transfer learning (STL). First, the data is divided into source and target domains, and a three-dimensional tensor containing spatial, feature, and temporal dimensions is constructed to achieve sliding-window-based icing prediction. Second, the model is built by combining PINN and STL. In the source domain branch, a PINN is trained to guide the model using physics knowledge. In the self-migration branch, multi-kernel maximum mean discrepancy is applied for domain adaptation. In the target domain branch, knowledge distillation is used to transfer physics knowledge from the expert model to the trained model. Finally, historical icing data from transmission lines in Guangxi province are used for simulation. Through interpretability analysis, the influence of meteorological, mechanical, line, and icing factors on line icing is revealed. Results show that compared to traditional data-driven models, the PKG-ISTL model improves prediction accuracy by 47.69%, verifying its effectiveness in small-sample scenarios. |
| Key words: icing prediction physics-informed neural network self-transfer learning knowledge distillation three- dimensional tensor |
