Real-Time Electromagnetic Field Prediction for 252kV GIS Using Integrated DNN and FEM Models

Gas-insulated switchgear (GIS) is an essential core component of modern power systems. Accurate prediction of the electromagnetic field distribution inside GIS is critically important for equipment design, performance optimization, and operational safety. A real-time electromagnetic field prediction model for GIS equipment plays a key role in helping power plant operators monitor the health of GIS and carry out maintenance effectively. Traditional numerical simulation methods, such as the finite element method (FEM), can provide high-precision electromagnetic field distribution results. However, their high computational and time costs have long been a barrier to real-time analysis and rapid iterative design. To address this issue, this study proposes a deep neural network (DNN) model framework based on FEM to achieve rapid real-time prediction of electromagnetic fields in GIS. First, an electromagnetic field database is generated under typical operating parameters using FEM simulations. This database is then used to train the DNN model to capture the mapping between GIS operating parameters and electromagnetic field distribution. FEM simulation results show that GIS operating parameters—such as operating current and insulation material properties—have a significant impact on the electromagnetic field distribution. More importantly, the DNN can reduce the prediction time from several hours (required by traditional methods) to just seconds, while maintaining high consistency with FEM results. The DNN model is four orders of magnitude faster than FEM and shows great potential for engineering applications.