| 基于FedGNN的10KV开关柜用氧化锌避雷器阀片热冲击裂纹扩展路径演化分析 |
投稿时间：2025-12-12 修订日期：2026-01-06  点此下载全文 |
| 引用本文： |
| 摘要点击次数: 32 |
| 全文下载次数: 0 |
|
|
| 中文摘要:为解决因数据孤岛导致的模型泛化性差问题，并精准捕捉裂纹沿阀片晶界扩展的非线性演化特征，研究基于FedGNN的10KV开关柜用氧化锌避雷器阀片热冲击裂纹扩展路径演化分析方法。在联邦学习框架下，客户端利用近场动力学算法，构建氧化锌避雷器阀片热冲击裂纹扩展数据集，并将其转换为客户端的本地图结构数据，建立局部图神经网络（GNN）模型，精准捕捉裂纹沿阀片沿晶界扩展的非线性演化特征，预测裂纹扩展路径，实现演化分析，并采用梯度下降法更新局部GNN模型参数；中央服务器采用基于强化学习的FedGNN客户端动态聚合策略，筛选并聚合优质客户端的局部GNN模型参数以更新全局GNN模型，解决因数据孤岛导致的模型泛化性差问题。实验证明：该方法可有效构建阀片热冲击裂纹扩展数据集，并预测氧化锌避雷器阀片热冲击裂纹扩展路径，完成扩展路径演化分析；在不同雷电热冲击电流幅值下，该方法演化分析的希尔不等系数最大值仅为0.05，表明其具有极高的分析精度。 |
| 中文关键词:联邦学习 图神经网络 10KV开关柜 氧化锌 避雷器阀片 热冲击裂纹 |
| |
| Analysis of the evolution of thermal shock Crack Propagation Path of Zinc Oxide Arrester Valve Plates for 10KV switch Cabinets Based on FedGNN |
|
|
| Abstract:To address the issue of poor model generalization caused by data silos and accurately capture the nonlinear evolution characteristics of crack propagation along the valve plate and grain boundaries, a method for analyzing the evolution of thermal shock crack propagation paths of valve plates for 10KV switch cabinets based on FedGNN is studied. Under the framework of federated learning, the client utilizes the near-field dynamics algorithm to construct a dataset of thermal shock crack propagation of zinc oxide arrester valve plates, and converts it into the client's local graph structure data. A local graph neural network (GNN) model is established to accurately capture the nonlinear evolution characteristics of crack propagation along the grain boundaries of the valve plates, predict the crack propagation path, and achieve evolution analysis. And the gradient descent method is adopted to update the local GNN model parameters; The central server adopts a FedGNN client dynamic aggregation strategy based on reinforcement learning, screening and aggregating the local GNN model parameters of high-quality clients to update the global GNN model, thereby solving the problem of poor model generalization caused by data silos. Experiments have proved that this method can effectively construct a dataset of thermal shock crack propagation of valve plates, predict the thermal shock crack propagation path of zinc oxide arrester valve plates, and complete the evolution analysis of the propagation path. Under different amplitudes of lightning thermal shock current, the maximum value of the Hill inequality coefficient in the evolution analysis of this method is only 0.05, indicating that it has extremely high analytical accuracy. |
| keywords:Federated learning Graph neural network 10KV switch cabinet Zinc oxide Lightning arrester valve plate Thermal shock crack |
| 查看全文 查看/发表评论 下载pdf阅读器 |
位访问者,今日访问
次