A Novel Deep Neural Network Framework for State Evaluation and Fault Diagnosis in Distribution Station

With the growing demand for reliable power supply from smart grid infrastructures, the assessment of operational status and fault diagnosis in distribution substations is becoming increasingly important. Traditional methods often fall short in providing real-time and accurate analysis, necessitating advancements in intelligent monitoring systems. To address this gap, a deep neural network framework, called SFCED, is proposed specifically for real-time state evaluation and fault diagnosis in distribution substations. This framework uses signature matrices to capture and represent correlations within multivariate time-series data, enabling a comprehensive understanding of distribution substation dynamics. SFCED integrates a fully convolutional encoder-decoder architecture, enabling the extraction of deep features from data and accurate reconstruction of system states. The empirical results, obtained from extensive comparisons with conventional CNN and LSTM Encoder-Decoder models, confirm the effectiveness of SFCED, particularly in achieving a high balance between precision and recall. The research demonstrates the practical applicability of SFCED in real-world scenarios, offering significant improvements in maintenance and operation efficiency for intelligent distribution substations.