Objective In response to the goals of carbon peaking and carbon neutrality, alongside the intelligence requirements of the new power system, a quadruple fusion algorithm based on digital twin and generative AI, named DRL-MLP-OBS-GAN (Deep Reinforcement Learning- Multi-Layer Perceptron- Optimal Brain Surgeon -Generative Adversarial Network) is proposed.

Method First, a digital twin model of the wind power system was constructed, enabling real-time mirroring of physical entity's states. Subsequently, GAN was deployed to emulate extreme fault scenarios under multifaceted disturbances, thereby enriching data diversity and the generalization capability of the model. Simultaneously, MLP-OBS optimized static feature extraction and model light weighting were combined, while the dynamic decision-making mechanism of DRL was introduced to implement adaptive fault dispatch strategies. Moreover, an innovative incorporation of Spatial Temporal Graph Convolutional Networks (ST-GCN) was employed to decipher the topological relationships of wind turbines and fault propagation paths, thereby enhancing multi-loop collaboration efficiency.

Result In the practical application at a wind farm situated in Guangdong, the algorithm achieves a prediction accuracy of up to 97.6% in 521 fault tests, with a response time of 0.62 s, generates an annual economic benefit of approximately 643 thousand yuan.

Conclusion Experimental results indicate that the algorithmic provides a standardized paradigm for fault dispatch in new power systems through the integration of virtual and physical elements, significantly enhances the resilience of power systems and facilitates the green transformation and intelligent upgrade of the power industry.