Circuit breaker fault diagnosis based on GWO-optimized CNN-LSTM

Abstract

In response to the current issues of low accuracy in circuit breaker fault diagnosis and the imbalanced distribution of fault feature samples, this paper aims to propose a new method for high-accuracy circuit breaker fault diagnosis. First, the Synthetic Minority Oversampling Technique (SMOTE) is introduced to expand the sample size of minority fault features, thereby effectively balancing the dataset; Second, a Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) model is constructed as the core foundation for feature extraction and fault classification; simultaneously, to address the challenges of hyperparameter tuning in the CNN-LSTM model-which significantly impacts diagnostic performance-this paper employs the Grey Wolf Optimization (GWO) algorithm to perform global optimization of its key hyperparameters, ultimately establishing the GWO-CNN-LSTM fault diagnosis model. Experimental results show that the comprehensive fault diagnosis accuracy of the proposed method for circuit breakers reaches as high as 99.17%; Compared to the DBO-CNN-LSTM model optimized using the Dung Beetle Optimization (DBO) algorithm and the WOA-CNN-LSTM model optimized using the Whale Optimization Algorithm (WOA), the diagnostic accuracy of this method has improved by 5% and 5.84%, respectively. The GWO-CNN-LSTM model proposed in this paper effectively addresses the challenge of high-precision classification under conditions of extreme data imbalance, demonstrating significant theoretical research value and practical engineering application potential in the field of high-voltage circuit breaker condition monitoring and fault detection.