Aiming at the issue of difficulty in effectively warning of faults in the pitch system of wind turbines, which affects the operational efficiency and reliability of wind power equipment under complex weather conditions, this study proposes a fault warning model combining Bidirectional Gated Recurrent Unit (BiGRU) and multi-head attention mechanism. By extracting key feature variables and utilizing BiGRU to learn the characteristics of SCADA data from wind turbines, the proposed model’s ability to capture feature information is enhanced. Additionally, the introduction of Cycle Network (CycleNet) and Crown Porcupine Optimization (CPO) algorithm optimizes the model’s prediction capabilities. The research results indicate that this model can effectively improve the accuracy of fault prediction in wind turbines, achieve fault warning, reduce downtime, and lower maintenance costs.