Development of a real-time fault detection method for electric power system via transformer-based deep learning model

Abstract

The real-time detection of power quality disturbances (PQDs) in electrical power systems (EPSs) is crucial for prompt actions to protect EPSs from cascade damages that can cause equipment failures, system downtime, and economic losses. The complexity of EPSs makes it challenging to detect the type and location of PQDs accurately. Many previous studies demonstrated that deep learning is a good tool for PQD detection, but the real-time capability of deep learning-based PQD detection has yet to be achieved. In this study, we proposed a voltage signal segmentation approach to use as an input of transformer-based deep learning models. To demonstrate the capability of the proposed method, synthetic voltage signals were prepared from the IEEE 9-bus system with four fault conditions using the PSCAD/EMTDC program. Segmented voltage signals, sampled every 1.67 ms, were used as the input of the deep learning models, and it successfully classified the type and location of PQDs, demonstrating the real-time capability (within 1.67 ms) of the proposed method. Additionally, we showed that the transformer-based model, trained using data obtained from just three locations, achieved high accuracy in detecting PQDs. We also demonstrated that the transformer-based model outperformed convolutional neural networks, which are conventional deep learning models for PQD detection. In conclusion, the suggested data segmentation approach with a transformer-based deep learning model opens up new possibilities in real-time PQD detection.