In the context of constructing digital power grid, there has been significant attention on the methods to accurately and promptly obtain the physical field information of power equipment. The numerical methods, such as finite element analysis (FEA), are limited by offline computation and cannot meet the safety and timeliness requirements of the power grid. In this letter, a method based on deep neural networks (DNN) is proposed for rapidly predicting the magnetic field distribution of reactors. After training on magnetic field data generated by FEA simulation, the DNN takes the reactor current value as input and gets the simulation results processed using principal component analysis (PCA) within 1 s. The trained DNN has a mean absolute percentage error (MAPE) of 0.012% in predicting the magnetic field distribution of reactor. This method demonstrates the viability of replacing traditional simulation methods with DNN to expand the applications of digital twin in the power grid.

中文翻译：

---

基于深度神经网络的反应堆磁场仿真

在建设数字电网的背景下，如何准确、及时地获取电力设备物理场信息的方法备受关注。有限元分析（FEA）等数值方法受限于离线计算，不能满足电网的安全性和时效性要求。在这封信中，提出了一种基于深度神经网络 (DNN) 的方法来快速预测反应器的磁场分布。在对 FEA 模拟生成的磁场数据进行训练后，DNN 将电抗器电流值作为输入，并在 1 秒内获得使用主成分分析 (PCA) 处理的模拟结果。经过训练的 DNN 在预测反应堆磁场分布时的平均绝对百分比误差 (MAPE) 为 0.012%。