The dynamic equivalent modeling of new energy stations is of great significance for the safe and stable operation of the power system under high penetration of new energy. Facing the challenge of dynamic equivalent modeling for doubly-fed wind farms, which relies on specific disturbances and thus struggles to acquire highly universal equivalent models, a dynamic equivalent modeling method for doubly-fed wind farms based on residual combination neural networks is proposed. Firstly, the mathematical model of wind turbines is simplified into a set of equations, with the input being the wind speed at the measurement tower and the voltage at the point of common coupling (PCC), and the output being the active and reactive power of the wind farm.Then, a residual combination neural network was constructed, comprising a feature memory layer with long short-term memory (LSTM) network units at its core, an information flow acceleration layer with residual-gate recurrent units (Res-GRU) at its core, and a data relationship mapping layer based on fully connected-batch normalization-rectified linear unit (FC-BN-ReLU). This network was employed to perform equivalent substitutions for the simplified equations of wind turbine units. Subsequently, a genetic algorithm (GA) is used to optimize the main parameters of the three components. Finally, taking a certain wind farm as an example, the equivalent modeling effect of the proposed network is compared with that of different methods. The results show that the equivalent accuracy of the proposed network is higher than that of traditional methods, and it can better fit the power response characteristics of the wind farm.