A Fault Feature Extraction Method for DC–DC Converters Based on Automatic Hyperparameter-Optimized 1-D Convolution and Long Short-Term Memory Neural Networks

To improve the diagnostic performance rate of parameter faults in dc–dc converters, obtaining the parameter fault characteristics of dc–dc converters is crucial. This study combines a 1-D convolutional neural network (1DCNN) and long short-term memory (LSTM) with automatic hyperparametric optimization and proposes deep learning models for time-series signal pattern recognition to achieve the dc–dc converter fault diagnosis. First, the 1-D original voltage signal of the dc–dc converters is preprocessed to obtain the labeled sample dataset. The corresponding training set and the test set are obtained. Then, according to the accuracy of the diagnosis, the optimal hyperparameter 1DCNN-LSTM diagnosis model is obtained from an optimization algorithm and the training sets. The model is applied to the test sets to perform feature extraction. Finally, a softmax classifier is used to evaluate the performance of the proposed method. Experimental results show that the proposed method outperforms the classic deep learning and traditional machine learning algorithms. In addition, the evaluation of the method under noise conditions is more challenging and practical.