An Integrated Physics‐Informed Deep CNN and Adaptive Elite‐Based PSO‐Catboost for Wind Energy Systems Fault Classification

ABSTRACT Intelligent fault diagnosis in wind energy systems requires accurate identification of faults since the annual maintenance cost can lead to substantial financial losses. Also, effective wind turbine fault diagnosis of critical fault types is essential, despite data discrepancies caused by unpredictable environmental conditions and human factors. This paper introduces a method combining deep learning with an optimized categorical boosting (CatBoost) model to improve fault classification using imbalanced SCADA data in wind energy systems. Our approach uniquely integrates t‐distributed stochastic neighbour embedding (t‐SNE) representations of the resampled SCADA data and its deep learning features extracted using a 1D physics‐informed deep convolutional neural network (PDCNN) with combined loss functions, namely, standard categorical cross‐entropy loss, deviation penalty loss and non‐negativity loss. Additionally, we introduce a framework for optimizing a categorical boosting (CatBoost) classifier using adaptive elite particle swarm optimization (AEPSO). The effectiveness of the proposed framework is validated with multiple recently developed deep learning models using highly imbalanced SCADA datasets. Experimental results demonstrate superior diagnostic performance, achieving higher accuracy and robustness compared to existing methods. This study aims to contribute an advanced methodology for wind turbine fault diagnosis by introducing a comprehensive framework that combines advanced deep learning and gradient boosting techniques to handle the complexities of imbalanced data and improve diagnostic reliability.