New data-driven approach to bridging power system protection gaps with deep learning

Protection is a critical function in power systems to avoid equipment damage, maintain personnel safety, and support system reliability. The existing protection technology that mainly relies on protective relays is not capable of providing 100% protection for all system events. Whenever protective relays fail to respond or respond incorrectly, it leads to the protection gap. In this paper, a new data-driven approach is proposed to complement the traditional protection technology so that the faulty conditions can be adequately distinguished from the transients resulted from normal operations. The proposed approach combines convolutional neural network and long short-term memory (CNN-LSTM) to develop a deep neural network that achieves the invariance in data translation and captures the temporal correlation of input data in time series. It can accurately detect the system faults with the variations and noises in the input data. The use of CNN-LSTM eliminates the complicated and often manual feature extraction procedure that is commonly required by conventional data-driven approaches. In order to address the issue of insufficient training data in practice, a transfer learning method is also applied to facilitate the future practical applications. The efficacy of the proposed data-driven approach is tested for the protection gaps with respect to the transmission line high-impedance fault and transformer inter-turn fault, respectively. The extensive study results demonstrate that the proposed approach can effectively bridge the protection gaps in power system operations.