Objective To address the limitations of traditional methods for low-frequency oscillation LFO identification in power systems under nonlinear and complex operating conditions a novel deep learning-based identification model is proposed. Its aim is to achieve accurate identification and effective suppression of oscillation modes thereby ensuring the safe and stable operation of power systems. Methods An identification model integrating a deep belief network DBN with a Softmax classifier was constructed. The DBN was employed to extract deep-level features from operational data through layer-wise pre-training and overall fine-tuning while the Softmax classifier performed the final mode classification. Subsequently the Prony method was incorporated for modal parameter estimation. Building upon this an oscillation trend prediction model and a parameter adjustment mechanism for power system stabilizers PSS were established forming a complete closed-loop framework encompassing ?? Identification-Estimation-Prediction-Control . Results The proposed model demonstrated excellent adaptability under various complex operating conditions achieving a significant improvement in identification accuracy. This validated the effectiveness of combining deep learning with traditional algorithms for tackling LFO problems. Additionally quantitative results on identification precision and suppression effectiveness were obtained. Conclusion In the context of diagnosing and controlling oscillation modes in practical power systems the method proposed in this study provides a novel and efficient solution. It holds significant engineering application value for enhancing power system stability in the future.