A Physics-Informed Neural Network Approach for Fault Diagnosis and Protection in Multi-Source Microgrids

Abstract

This paper presented a novel framework based on Physically Informed Neural Networks (PINN) for the protection and monitoring of three-phase microgrids. In order to generate data, a microgrid including distributed generation resources was modeled in the Simulink/Matlab environment and voltage and current data were extracted. Then, the obtained data was transferred to the Google Colab environment, and the proposed algorithm was implemented in the Python platform. The distinctive feature of this approach compared to conventional intelligent methods is the combination of physical laws of the power system with simulated data, which allows for deeper learning of network dynamics and more accurate identification of fault states. The evaluation results show that the proposed model is capable of detecting single-phase, two-phase and three-phase faults even in noisy conditions with an overall accuracy higher than 97%. Furthermore, comparison with conventional machine learning algorithms such as SVM, KNN, DT, RF and ANN showed that the proposed method performs better in terms of stability, accuracy and noise immunity. Therefore, the developed framework can be used as an efficient tool to enhance the stability, reliability and protection of modern microgrids.