Electromagnetic Interference Effect Assessment Under Measuring Testability Limitation Based on Physics-Informed Neural Network and Gaussian Process Regression

The transient electromagnetic interference in practical existing engineering have measuring testability limitation. This limitation makes it difficult to assess electromagnetic effect. A multidimensional physical information based assessment method via physics-informed neural network (PINN) and Gaussian process regression (GPR) is proposed in this article. First, the physical topology of the disturbed equipment under electromagnetic interference is analyzed, and the partial differential equation of the space electric field is derived from the measurable electric quantity. Then, the electric field near the disturbed equipment is predicted by PINN which can characterize the physical information. Finally, combining calculated electric field data and measured magnetic field data, a two-dimensional GPR regression is used to assess the electromagnetic effect with multiple inputs and outputs, and the probabilities under each state are then calculated based on the assessed state. Expanded verification is conducted by combining different electromagnetic complexity calculation methods, proving the applicability and effectiveness of the proposed assessment model.