This lecture introduces a deep learning–based surrogate modeling approach for power electronic converters, developed exclusively from external measurements at the point of common coupling (PCC). The objective is to approximate the dynamic behavior of converters and multi-converter systems without relying on detailed internal models, enabling stability assessment under realistic confidentiality constraints.The methodology relies on admittance-based representations combined with machine learning techniques. First, a surrogate model isdemonstrated for a single grid-connected converter. Using datasets generated from electromagnetic transient simulations, classification models identify the implemented control mode (Grid-Following or Grid-Forming) and control structure. A dedicated 1D convolutional neural network achieves highly accurate classification. Regression models based on deep neural networks then estimate the control parameters, allowing the surrogate to faithfully reproduce the converter admittance over a wide frequency range. The presentation is subsequently extended to multi-converter systems, with a focus on wind farms. A novel method is showcased to estimate the proportion of Grid-Forming and Grid-Following converters from the global admittance measured at the PCC. To mitigate topological uncertainty, a denoising autoencoder is used to extract the net converter dynamics, followed by an LSTM-based estimator achieving a mean absolute percentage error below 5%. An equivalent surrogate model of the wind farm is finally obtained by aggregating converters and estimating an equivalent network representation.