A new protection scheme for loss of excitation based on CNN

If a loss of excitation (LOE) occurs in a synchronous generator, this leads to a reduction in the output voltage and to a considerable absorption of reactive power from the system. This can cause severe damage to both the generator and the power system. Conventional methods often suffer from low speed and are susceptible to disturbances from events such as stable power swings (SPS), three-phase faults or partial loss of excitation (PLOE) on the power system. It is important that the relay can accurately distinguish between these phenomena and LOE. In this paper, a novel intelligent method is proposed to improve the speed and sensitivity of LOE detection and protection of synchronous generators. The proposed scheme utilizes a convolutional neural network (CNN) in a Python environment, eliminating the need for separate signal preprocessing and feature extraction. The simulation of the example system and the protection system was carried out in the MATLAB/Simulink software environment. By using CNN, the proposed scheme reduces the detection time to 0.96 s and achieves an overall accuracy of 96.47% in the presence of all the mentioned phenomena. It also achieves an overall accuracy of 99.12% regardless of PLOE. It improves the efficiency and accuracy of LOE detection and effectively differentiates it from SPS, three-phase faults, and partial LOE. In addition, the CNN method recognizes multiple phenomena in the power system phenomena and improves accuracy and speed. This tailored approach highlights potential advances in power system protection and monitoring and illustrates the advantages of a 1-dimensional convolutional neural network (1D CNN) for analyzing time series data in this context.