上海交通大学学报

上海交通大学学报

• • 上一篇    下一篇

基于深度算子网络的变压器套管将军帽多物理场快速计算方法

  

  1. 1. 上海交通大学 电气工程学院，上海 200240；

    2. 中国南方电网有限责任公司超高压输电公司电力科研院，广州 510620
  • 作者简介:柏津（2002—），硕士生，从事输变电装备多物理场仿真研究

Fast Multi-Physics Field Computation Method for Top Flange of Transformer Bushing Based on Deep Operator Network

  1. 1. School of Electrical Engineering, Shanghai Jiao Tong University, Shanghai 200240, China;

    2. Electric Power Research Institute of EHV Power Transmission Company of China Southern Power Grid Co., Ltd., Guangzhou 510620, China

PDF (PC)

129

摘要： 针对目前电力装备多物理场耦合仿真效率低及现有快速计算方法误差较大的问题，提出一种基于深度算子网络(Deep Operator Network, DeepONet)的电热力耦合多物理场快速计算方法。以110 kV套管为例，对有限元仿真进行拉丁超立方采样以生成数据集，结合多任务DeepONet框架，利用梦境优化算法(Dream Optimization Algorithm，DOA)进行超参数寻优，实现变压器套管将军帽应力场、温度场及位移场分布的高精度快速计算。该方法单次求解耗时仅0.005s，应力场、温度场和位移场准确率分别为90.21%、99.87%和99.09%；相较于U-net和DNN代理模型，其在多项评价指标及鲁棒性方面均有显著提升。最后通过实验验证了模型在实际工况下的适用性。该模型为电力装备结构优化、状态感知、多物理场重构等数字孪生场景提供了可靠工具。

关键词:

"> 变压器套管, 多物理场快速计算, 深度算子网络, 梦境优化算法, 代理模型

Abstract: Aiming at the current low efficiency of multi-physical field coupling simulation of electric equipment and the large error of the existing fast calculation method, a fast calculation method of electric-thermal coupling multi-physical field based on Deep Operator Network (DeepONet) is proposed. Taking 110 kV bushing as an example, Latin hypercubic sampling of finite element simulation is carried out to generate a data set, combined with multi-task DeepONet framework, and hyperparameter optimization is carried out by using Dream Optimization Algorithm (DOA) to realize high-precision and fast computation of the distribution of top flange stress, temperature, and displacement fields of the transformer bushing. This method only takes 0.005s for a single solution, and the accuracy of stress field, temperature field, and displacement field are 90.21%, 99.87%, and 99.09%, respectively; Compared to U-net and DNN proxy models, it has significantly improved in multiple evaluation metrics and robustness. Finally, the applicability of the model in real working conditions is verified by experiments. The model provides a reliable tool for digital twin scenarios such as power equipment structure optimization, state sensing, and multi-physical field reconstruction.

Key words: transformer bushing, multi-physics fast computation, deep operator network (DeepONet), dream optimization algorithm (DOA), surrogate model