Wind Power Turbine Fault Detection using Hybridized Convolutional Neural Network and Autoencoder

Objective: This study aims to design and develop a hybrid border collie-based satin bower (BCSB) optimization algorithm for short circuit fault detection in the PMSG-based wind turbine system. Methods: This paper detects the stator short circuit fault that occurs within its phase turns and phase to ground, optimally with a hybrid classifier. The hybrid optimized classifier includes an optimal convolution neural network (CNN) and an auto encoder (AE). The hybridized classifier has been used for the optimal detection of faults using a novel BCSB optimization algorithm trained with short-circuit fault data. In this approach, a cascaded hybridization is performed, where the update process of SBO is serially linked with the update process of BCO. This creates a dual exploration process, allowing the search procedure to effectively determine the optimal weights for the CNN. The efficiency of the novel optimized method has been validated using Simulink in terms of error analysis, and the proposed method has been evaluated using the convergence analysis. Results: The proposed method outperforms traditional algorithms in terms of fault detection accuracy. Specifically, the root mean squared error (RMSE) for the [Formula: see text] model at [Formula: see text] is 0.33551, which is 4.73%–6.08% lower than existing methods like WO, DFO, GWO, SBO and BCO. Additionally, the mean squared error (MSE) is reduced by 0.67% to 12.75%, and the mean absolute percentage error (MAPE) shows a reduction of 36.63% to 2.07% compared with the traditional approaches. Conclusions: The result shows that the proposed method hybrid [Formula: see text] model optimized by the BCSB algorithm proves to be a robust and efficient solution for the early detection and diagnosis of faults in wind turbine generators, offering valuable insights for enhancing the reliability of WPT systems.