Abstract: The transmission characteristics of electromagnetic signals in vertical transition structures are highly complex, making it particularly important to study the matching between different multilayer boards. As the complexity of the transition structure increases, the time required for electromagnetic simulation also significantly rises. To address this issue, this paper presents a parametric modeling approach for the electromagnetic behavior of vertical transitions based on neural networks and the residue-pole method. In this study, a fully connected neural network is used to establish a mapping relationship between physical parameters and the electromagnetic behavior of vertical transition structures, allowing for precise capture of their complex nonlinear characteristics. For S-parameter data with varying frequencies, the pole-residue method is introduced for preprocessing, representing the frequency domain data in terms of poles and residues, thereby effectively reducing the dimensionality and complexity of the data. Experimental results show that the average error between the neural network model’s predictions and actual data at different frequencies is less than 1dB, indicating that the proposed method can accurately predict the electromagnetic behavior of vertical transition structures. Compared to traditional electromagnetic simulations, the neural network model offers extremely high prediction efficiency, providing a theoretical basis for modeling complex electromagnetic systems and optimizing subsequent circuit designs.