Crested Porcupine Optimized Long Short-Term Memory based Multi Fault Diagnosis of Wind Turbines

Wind turbines (WT) operate continuously under dynamic and harsh environmental conditions making them prone to failure. This necessitates a reliable and robust fault diagnostic system to monitor and ensure their safe and uninterrupted operation. Long Short-Term Memory (LSTM), commonly used for WT fault diagnosis, often has its diagnostic performance hinging on the optimal hyper parameters selection. To address these challenge, Crested Porcupine Optimization (CPO) algorithm with adaptive hyperparameter optimization is proposed. CPO's adaptive optimization improves the LSTM’s temporal fault pattern learning, thereby improving the generalization capability, and the diagnostic accuracy of the proposed CPO-LSTM diagnostic model. To enhance the computational efficiency and to avoid overfitting, ReliefF-PCA-based dimensionality reduction was applied to select the most informative SCADA features. The proposed CPO–LSTM model was validated using real-world SCADA data collected from a 2 MW wind turbine and its practical deployment aspects and hardware requirements, are discussed to demonstrate real-world applicability. The classification accuracy and robustness of the proposed model were analyzed by comparing the experimental results with those of the LSTM models tuned with traditional metaheuristic-based fault diagnostic models. The results indicate that the proposed CPO–LSTM framework is highly accurate (with an accuracy of 98.4%) and more robust, making it suitable for practical WT condition monitoring..