Abstract: To enhance the design, manufacturing, and operational reliability of power equipment, the finite element method (FEM) is widely employed to analyze electromagnetic field distributions under varying operating conditions. However, the computational expense associated with FEM-based parametric analyses has led to an increasing reliance on deep learning models in engineering applications. Traditional convolution-based models, such as U-net convolutional neural networks, often suffer from the loss of high-frequency information during pooling operations and fail to incorporate positional coordinates effectively. This paper introduces a rapid computational approach for electromagnetic field analysis based on the deep operator network (DeepONet). The dataset is initially generated through simulations or experiments, and the problem of inconsistency in grid point numbers under varying boundary conditions is solved by employing uniform complementary points. Fourier transformations are applied to the input data to enhance the neural network's capability for extracting high-frequency features. Within the DeepONet framework, multiple variable parameters and positional coordinates serve as inputs, while outputs include magnetic vector potential or electric potential, with derived physical quantities such as magnetic induction strength and electric field strength calculated via automatic differentiation. A simplified two-dimensional arithmetic example is presented to evaluate the impact of different differential constraints on network performance, followed by an extension to three-dimensional cases. The results demonstrate that the proposed method achieves computational errors below 3% while significantly reducing computation time compared to FEM, enabling second-level computations. Additionally, higher accuracy and greater ease of post-processingcan be obtained compared to the U-net method. This approach is particularly suitable for applications requiring extensive parametric computations during the design phase and for real-time applications such as equipment condition monitoring and digital twin implementations.