Grid fault diagnosis based on the deep pyramid convolutional neural network

Traditional power grid fault diagnosis methods rely on manual experiences to handle massive amounts of alarm information, have complex modeling processes and insufficient generalization abilities, and lack direct diagnostic research on the alarm information text. Therefore, we propose an intelligent fault diagnosis method based on the deep pyramid convolutional neural network (DPCNN), where we build an end-to-end fault classification model and a key information extraction model to directly mine the implicit fault features from the alarm information text to achieve accurate classification of fault types and rapid location of faulty equipment. We performed comparative experiments to show that the proposed method performs well in complex power grid scenarios and noisy data environments; the highest fault classification accuracy achieved with this approach was up to 100%, and we could effectively identify multiple fault types, such as simple faults, switch operation failure, and protection operation failure. In addition, we integrated the temporal-sequence-prioritized faulty equipment identification strategy with the proposed method to further improve the fault location accuracy. A case study verification was also performed, which shows that our method has a fault recognition rate of up to 99.5% and can achieve 98.7% accurate positioning after one-by-one elimination through the identification strategy to significantly reduce manual intervention and have high applicability in actual power grids.