Multi-fidelity methods are commonly used to construct surrogate models for complex systems when high-fidelity simulations are computationally expensive or experimental data are difficult to obtain. The major challenge in such methods is the generation of training data, as high-fidelity electromagnetic simulations can require significant computational time. Multi-fidelity modeling addresses this challenge by combining a limited number of high-fidelity simulations with a larger set of computationally inexpensive low-fidelity models that capture the same underlying physical mechanisms with reduced accuracy. In electromagnetic interference and compatibility (EMI/EMC) analysis, multi-fidelity based modeling is particularly challenging because system behavior is often dominated by resonances. Simplifications in low-fidelity models can lead to discrepancies in resonance frequency, amplitude, and sensitivity to design parameter variations, resulting in weak correlation between low- and high-fidelity responses. This mismatch degrades prediction accuracy and may introduce non-physical resonances in conventional multi-fidelity surrogate models. To address these limitations, this work proposes a multi-fidelity surrogate modeling framework based on autoencoders (AE) for nonlinear dimensionality reduction. The autoencoder is trained using low-fidelity simulation data to learn a compact latent representation of frequency-domain system responses. This latent representation is subsequently used together with a limited set of high-fidelity simulation data to train a regression model that captures the relationship between low- and high-fidelity responses. The proposed approach is validated using an EMC filter circuit, with three-dimensional full-wave electromagnetic simulations as high-fidelity data and network-based simulations as low-fidelity data. The results demonstrate improved prediction accuracy of resonant behavior while reducing the number of required high-fidelity simulations, enabling efficient surrogate modeling for EMC analysis of power electronic systems.