Abstract: The accurate acquisition of control parameters for doubly-fed induction generators (DFIGs) presents a significant challenge, which impedes the precise establishment of DFIG simulation models and poses substantial obstacles to the simulation analysis of large-scale wind power integration into power systems. To address this challenge, this study introduces a new control parameter identification approach for DFIGs that integrates recursive feature elimination (RFE) with bidirectional long short-term memory (BiLSTM) neural networks. First, the dynamic response measurement dataset of a real DFIG controller is acquired through the RT-LAB platform. Secondly, this dataset is utilized to populate the Simulink simulation identification model of the DFIG. Subsequently, the adaptive random test strategy is employed to enhance the performance of the genetic algorithm, which is then used to generate the training dataset for the BiLSTM. The RFE-BiLSTM model is utilized to identify the control parameters of the DFIG and obtain the key control parameters. Finally, the efficacy of the proposed method is validated through practical examples. The results indicate that, in comparison to the long short-term memory (LSTM) and BiLSTM models, the RFE-BiLSTM model presented in this paper exhibits the lowest identification error across the entire time scale and in five specific segments. Additionally, it is found that the adaptability of the identification results to various operating conditions is superior.