Dynamic equivalent modeling of doubly-fed wind farms relies on specific disturbances, making it challenging to obtain universally applicable equivalent models. To address this issue, this paper proposes a data-driven approach for dynamic equivalent modeling of doubly-fed wind farms. Firstly, the mathematical model of the doubly-fed wind turbine units is simplified into a set of equations. Secondly, neural network components with similarity to these equations are selected and reasonably combined to construct a residual combination neural network. This network includes feature memory layers, information flow acceleration layers, and data relationship mapping layers, aimed at simplifying the detailed wind farm model equivalently. Subsequently, genetic algorithms are employed to optimize the main parameters of the three components in this combined network. A typical wind farm in Henan Province is used as an example for test analysis. The results show that the proposed method based on residual combination neural networks accurately characterizes the output response characteristics of wind farms, achieving higher modeling precision.