As the core power generation component in wind power systems

the Permanent Magnet Synchronous Generator (PMSG) can significantly impact the entire system upon failure

making timely fault diagnosis and identification crucial for system maintenance. Addressing current challenges in PMSG internal fault identification-such as scarce and imbalanced real fault data

complex internal signal processing

and insufficient deep fault feature extraction-this paper proposes a PMSG fault identification method driven by the integration of multi-dimensional Transformer algorithms and electromagnetic transient modeling (EMT). First

an electromagnetic transient equivalent model of PMSG covering five typical internal faults is constructed on the PSCAD/EMTDC platform

with magnetic flux as the core state variable. Integrated into a wind turbine under multiple operating conditions

it generates a comprehensive and balanced dataset for model training. Then

a multi-dimensional Transformer encoder is proposed

employing parallel time-frequency feature extraction and a gating mechanism for fault classification. Test results demonstrate that the proposed method achieves 99.8% identification accuracy with 60% faster training convergence than traditional models

exhibiting excellent identification performance.