<p>In the design optimization of electrical equipment, it is necessary to analyze magnetic field distribution characteristics under massive parameter combinations. Traditional finite element methods face exponential growth in computational time and resource consumption when dealing with increasingly sophisticated electrical equipment. To address this challenge, this study proposes a hybrid method combining finite element analysis and deep learning for magnetic field prediction in electrical equipment. The method achieves end-to-end intelligence from geometric modeling to prediction output through seamless integration of an adaptive magnetic field simulation system with deep learning models, employing an iterative data generation strategy that dynamically optimizes training data based on prediction errors. The study introduces hierarchical enhanced feature network, a deep learning model featuring hierarchical feature enhancement, which improves feature extraction capabilities through multi-directional sequence transformation and spatial state sequence modeling. Validation using cable, transformer, and motor models of varying complexity demonstrates that this method reduces prediction errors by 60.31% compared to U-Net and Swin Transformer models, paving new paths for deep learning applications in electrical equipment modeling, numerical analysis, and parameter optimization design.</p>