A modified BiLSTM network for minor fault identification of power system

Given the high dimensionality and intricate characteristics of power measurement data, conventional attack identification methodologies struggle to effectively detect faults and false attacks within power systems, thereby jeopardizing the secure and stable operation of the power grid. This study introduces a novel recognition framework, termed CNN BiLSTM + QAN, designed specifically for the identification of minor faults and attacks in power systems. The proposed network employs Convolutional Neural Networks (CNN) to extract salient features from fault signals and integrates the Quantum Attention Network (QAN) into a Bidirectional Long Short-Term Memory (BiLSTM) architecture, thereby constructing a BiLSTM + QAN recognition network that significantly enhances the characterization and recognition capabilities of localized fault signal features. To mitigate the influence of anomalous noise, the loss function is formulated using Cross Entropy Loss, and gradient-based optimization techniques are employed to achieve global convergence of the network, further augmenting its efficacy in detecting minor faults. Utilizing simulated data of minor faults from the IEEE 14-bus system, comparative experiments demonstrate that the CNN BiLSTM + QAN framework developed in this study effectively recognizes minor faults, achieving an impressive recognition accuracy of 98.34%.