Multi-Channel Convolutional Neural Network with Long Short-Term Memory for Power Grid Safety Monitoring

Power distribution grids are commonly installed in outdoor environments, making them susceptible to various environmental conditions. However, capturing the time dependencies of grid power based on augmented data is crucial for ensuring timely and sensitive monitoring. In this research, propose a Multi-Channel Convolutional Neural Network (MCCNN) with Long Short-Term Memory (LSTM) to integrate and maintain balanced standards of signals, current, and voltage. Feature fusion using Principal Component Analysis (PCA) identifies relevant features or characteristics from the data analysis that indicate safety-related events or anomalies in the grid. The CNN structure consists of convolutional layers, pooling layers, and fully connected layers. LSTM network, which is a variation of Recurrent Neural Network (RNN), is characterized by a directed cycle structure. This structure allows the network to transfer the output of a hidden layer to capture features and time-series information effectively. The proposed MCCNN-LSTM technique is evaluated on power grid data, achieving better accuracy 97.25%, precision 98.68%, recall 92.95%, and F1-score 95.72%, with an AUC of 97.05%. This approach proves to be more effective than existing methods such as the Semi-ProtoPNet Deep learning model (SPPD) with VGG19, Deep Neural Network (DNN), and Graph Convolutional Attention Network (GCAT).