融合物理信息与深度算子网络的变压器绕组热-流场快速计算方法

李家豪; 王青于; 刘鹏; 吴泽华; 程建伟; 曾强

doi:10.7652/xjtuxb202607021

您当前的位置：

首页 >

文章列表页 >

融合物理信息与深度算子网络的变压器绕组热-流场快速计算方法

更新时间：2026-07-09

- 融合物理信息与深度算子网络的变压器绕组热-流场快速计算方法
- A Fast Calculation Method for Thermal Flow Fields in Transformer Windings Based on PI-DeepONet
- 西安交通大学学报 2026年60卷第7期页码：228-238
- 作者机构：
  
  西安交通大学电工材料电气绝缘全国重点实验室，710049，西安
  中国南方电网有限责任公司南方电网科学研究院有限责任公司，510663，广州
  广东电网有限责任公司东莞供电局，523009，广东东莞
- 作者简介：
  
  作者简介：李家豪（1999—），男，博士生；
  王青于（通信作者），女，副教授，硕士生导师。
- 基金信息：
  
  广东省重点领域研发计划资助项目(2021B0101190003);南方电网有限公司总部科技资助项目(GDKJXM20222106)
- DOI：10.7652/xjtuxb202607021
  中图分类号： TM411
- 收稿：2025-09-01，
  
  纸质出版：2026-07-10
- 稿件说明：
移动端阅览
李家豪, 王青于, 刘鹏, 等. 融合物理信息与深度算子网络的变压器绕组热-流场快速计算方法[J]. 西安交通大学学报, 2026,60(7):228-238.

LI Jiahao, WANG Qingyu, LIU Peng, et al. A Fast Calculation Method for Thermal Flow Fields in Transformer Windings Based on PI-DeepONet[J]. Journal of Xi'an Jiaotong University, 2026, 60(7): 228-238.
李家豪, 王青于, 刘鹏, 等. 融合物理信息与深度算子网络的变压器绕组热-流场快速计算方法[J]. 西安交通大学学报, 2026,60(7):228-238. DOI： 10.7652/xjtuxb202607021.

LI Jiahao, WANG Qingyu, LIU Peng, et al. A Fast Calculation Method for Thermal Flow Fields in Transformer Windings Based on PI-DeepONet[J]. Journal of Xi'an Jiaotong University, 2026, 60(7): 228-238. DOI： 10.7652/xjtuxb202607021.

摘要

针对传统有限元方法在变压器绕组温度评估中计算成本高、求解速度慢、难以满足工程应用需求的问题，提出了一种基于物理信息深度算子网络的变压器稳态热-流场快速计算方法，可实现热-流场的高精度、高效率计算。该方法基于数值仿真获取不同工况下绕组热-流场的分布特征，构建高质量数据集；将深度算子网络结构与物理约束相结合，设计包含主干网络与分支网络的架构，引入热传导方程和流体连续性方程作为物理约束构建损失函数，实现对变压器绕组稳态热-流场的高精度预测。算例结果表明：在神经网络深度为7层、宽度为50、学习率为0.01、Tanh激活函数及Adam优化器的超参数配置下，所提方法对温度、压强及径向、轴向油流速度的预测平均绝对误差分别为0.1℃、5.03 Pa、0.0061 m/s、0.0036 m/s；95%以上节点相对误差低于0.7%，单次预测耗时0.12 s，较Fluent数值仿真加速约10000倍。所提方法在保持热-流场预测精度的同时显著提升计算时效，可支撑变压器热管理与数字孪生的快速仿真需求。

Abstract

To address the limitations of traditional finite element methods（FEM）in transformer winding temperature evaluation，such as high computational costs，slow solution speeds，and poor suitability for engineering applications，a fast calculation method for steady-state thermal flow fields in transformers is proposed based on a physics-informed deep operator network（PI-DeepONet）.This method achieves high-precision and high-efficiency calculation of thermal flow fields.The distribution characteristics of the thermal flow fields in transformer windings under various operating conditions were obtained via numerical simulations，and a high-quality dataset was constructed.By integrating the DeepONet architecture with physical constraints，a network comprising branch and trunk components was designed.The heat conduction and fluid continuity equations were incorporated into the loss function as physical constraints，enabling accurate prediction of the steady-state thermal flow field in transformer windings.Calculation case results demonstrate that under a hyperparameter configuration featuring a 7-layer depth，a 50-unit width，a learning rate of 0.01，a Tanh activation function，and the Adam optimizer，the mean absolute errors（MAE）for temperature，pressure，and radial/axial flow velocities are 0.1 ℃，5.03 Pa，0.0061 m/s，and 0.0036 m/s，respectively.Relative errors for over 95% of the nodes are below 0.7%.Furthermore，a single prediction takes only 0.12 s，representing a speedup of approximately 10000 times compared to Fluent-based numerical simulations. The proposed method significantly enhances computational efficiency while maintaining prediction accuracy of the thermal flow fields，supporting the rapid simulation requirements of thermal management and digital twin applications for transformers.

关键词

Keywords

references

国家能源局电力可靠性管理和工程质量监督中心. 2022年全国电力可靠性年度报告[EB/OL].(2023-10-18)[2025-05-13].https://prpq.nea.gov.cn/%20ndbg/9875.html.

江秀臣, 许永鹏, 李曜丞, 等.新型电力系统背景下的输变电数字化转型[J].高电压技术, 2022, 48(1):1-10.

Jiang Xiuchen, Xu Yongpeng, Li Yaocheng, et al. Digitalization transformation of power transmission and transformation under the back-ground of new power system [J].High Voltage Engineering, 2022, 48(1):1-10.

盛戈皞, 钱勇, 罗林根, 等.面向新型电力系统的数字化电力设备关键技术及其发展趋势[J].高电压技术, 2023, 49(5):1765-1778.

Sheng Gehao, Qian Yong, Luo Lingen, et al.Key technologies and development trends of digital power equipment for new type power system [J].High Voltage Engineering, 2023, 49(5):1765-1778.

吴泽华, 吴宝英, 赵林杰, 等.面向输变电设备数字孪生的多物理场正反演快速仿真关键技术综述[J].电网技术, 2024, 48(10):4215-4230.

Wu Zehua, Wu Baoying, Zhao Linjie, et al.Review of forward and reverse multi-field simulation technology for digital twin of electrical power equipment [J]. Power System Technology, 2024, 48(10):4215-4230.

李元, 王晨, 张冠军.传热学与电学问题的迁移类比研究:(2)工程实践[J].西安交通大学学报, 2021, 55(9):1-8.

Li Yuan, Wang Chen, Zhang Guanjun.Analogical transfer between heat transfer and electrology:(2)engineering practices [J].Journal of Xi'an Jiaotong University, 2021, 55(9):1-8.

张志宇, 寇家俊, 张重远, 等.基于本征正交分解和克里金代理模型的油浸式变压器绕组温度场快速计算方法[J].电工技术学报, 2025, 40(7):2215-2225.

Zhang Zhiyu, Kou Jiajun, Zhang Zhongyuan, et al.A fast calculation method for temperature field of oil immersed transformer windings based on proper orthogonal decomposition and Kriging surrogate model [J]. Transactions of China Electrotechnical Society, 2025, 40(7):2215-2225.

赵庆贤, 刘云鹏, 刘刚, 等.基于POD-DNN降阶模型的油浸式变压器绕组稳态温升快速计算方法[J].中国电机工程学报, 2025, 45(6):2423-2436.

Zhao Qingxian, Liu Yunpeng, Liu Gang, et al.Fast calculation of steady temperature rise of oil-immersed transformer windings based on POD-DNN reduced order model [J].Proceedings of the CSEE, 2025, 45(6):2423-2436.

Lu Lu, Jin Pengzhan, Pang Guofei, et al.Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators [J].Nature Machine Intelligence, 2021, 3(3):218-229.

魏蓉, 陈锦培, 仲林林.基于深度算子网络的电磁轨道发射速度趋肤效应的快速计算方法[J].电工技术学报, 2025, 40(5):1344-1354.

Wei Rong, Chen Jinpei, Zhong Linlin.A fast computational method for velocity skin effect of electromagnetic rail launch based on deep operator network [J]. Transactions of China Electrotechnical Society, 2025, 40(5):1344-1354.

张宇娇, 赵志涛, 赵常威, 等.基于深度算子网络的电磁场快速计算方法[J].高电压技术, 2025, 51(3):1484-1494.

Zhang Yujiao, Zhao Zhitao, Zhao Changwei, et al. Rapid calculation method for electromagnetic fields based on deep operator network [J]. High Voltage Engineering, 2025, 51(3):1484-1494.

Sasaki H, Iwata K, Sato T, et al.Prediction of motor characteristic maps via deep operator networks for topology optimization [J].IEEE Transactions on Magnetics, 2024, 60(12):1-5.

Xu Chen, Cao B T, Yuan Yong, et al.A multi-fidelity deep operator network (DeepONet)for fusing simulation and monitoring data:application to real-time settlement prediction during tunnel construction [J].Engineering Applications of Artificial Intelligence, 2024, 133(PB):108156.

Ahmed B, Qiu Yuqing, Abueidda D W, et al.Physics-informed deep operator networks with stiffness-based loss functions for structural response prediction [J].Engineering Applications of Artificial Intelligence, 2025, 144:110097.

Shukla K, Oommen V, Peyvan A, et al.Deep neural operators as accurate surrogates for shape optimization [J].Engineering Applications of Artificial Intelligence, 2024, 129:107615.

张振, 苏欣荣, 袁新.基于物理信息神经网络的气膜冷却湍流模型反演学习[J].动力工程学报, 2024, 44(9):1459-1465.

Zhang Zhen, Su Xinrong, Yuan Xin.PINN-based inversion learning of turbulence model for film cooling [J]. Journal of Chinese Society of Power Engineering, 2024, 44(9):1459-1465.

金立军, 王飞龙, 王金宇.基于物理信息网络的开关柜三维温升场域重建与缺陷热源反演[J].中国电机工程学报, 2026, 46(4):1685-1697.

Jin Lijun, Wang Feilong, Wang Jinyu.Three-dimensional temperature rise field reconstruction and defect heat source inversion of switchgear based on physical information network [J].Proceedings of the CSEE, 2026, 46(4):1685-1697.

张宇娇, 孙宏达, 赵志涛, 等.基于物理信息神经网络的电磁场计算方法[J].电工技术学报, 2024, 39(17):5251-5261.

Zhang Yujiao, Sun Hongda, Zhao Zhitao, et al.Electromagnetic field calculation method based on physical informed neural network [J].Transactions of China Electrotechnical Society, 2024, 39(17):5251-5261.

杨立.使用DeepONet模型求泊松方程解算子[D].成都:电子科技大学, 2024:16-21.

席梓严, 戴玉婷, 黄广靖, 等.基于深度算子神经网络的翼型失速颤振预测[J].力学学报, 2024, 56(3):626-634.

Xi Ziyan, Dai Yuting, Huang Guangjing, et al.Airfoil stall flutter prediction based on DeepONet [J].Chinese Journal of Theoretical and Applied Mechanics, 2024, 56(3):626-634.

尹掀浪, 苏晶晶, 张升,等.基于PI-DeepONet算法与稀疏测点数据的两类饱和软土固结行为预测[J].铁道科学与工程学报, 2025, 22(10):4542-4552.

Yin Xianlang, Su Jingjing, Zhang Sheng, et al.Prediction of two types of saturated soft soil consolidation behavior based on the PI-DeepONet algorithm and sparse measurement data [J].Journal of Railway Science and Engineering, 2025, 22(10):4542-4552.

黄子杰.基于深度算子网络的非饱和土壤水分运动模拟研究[D].杭州:浙江大学, 2024:17-24.

王非凡.基于深度算子网络的飞行器热防护不确定性量化分析[D].北京:华北电力大学, 2024:33-35.

陈思, 盛戈皞, 张立静, 等.基于绕组等效热参数的干式变压器温度场分布快速仿真方法[J].高电压技术, 2025, 51(1):97-109.

Chen Si, Sheng Gehao, Zhang Lijing, et al.Fast simulation method of temperature field distribution for dry-type transformer based on equivalent thermal parameters of windings [J].High Voltage Engineering, 2025, 51(1):97-109.

李元, 陈智秋, 刘志濠, 等.伴随网格自适应剖分优化变压器内部温度仿真模型计算效率方法[J].西安交通大学学报, 2026, 60(3):166-174.

Li Yuan, Chen Zhiqiu, Liu Zhihao, et al.Adjointbased adaptive mesh refinement method for optimizing computational efficiency of transformer internal temperature simulation models [J].Journal of Xi'an Jiaotong University, 2026, 60(3):166-174.

Wang Jiangyu, Peng Xingjie, Chen Zhang, et al.Surrogate modeling for neutron diffusion problems based on conservative physics-informed neural networks with boundary conditions enforcement [J].Annals of Nuclear Energy, 2022, 176:109234.

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

电子级超纯氮深冷空分流程节能优化研究

扇贝仿生密封腔室内超临界二氧化碳泄漏流湍动能耗散机理研究

语义区分和子图整合驱动的知识图谱表示学习方法

隧道内燃料电池汽车氢气泄漏扩散特性及影响因素

面向高海拔空冷发电机防晕性能演化规律的电热耦合研究