Investigation of different machine learning approaches for fault identification in microgrid

The growing adoption of microgrids in modern energy systems has underscored the critical need for robust protection mechanisms to ensure their reliability and stability. With the dynamic nature of microgrids resulting from the large integration of renewable energy sources, conventional fault prediction schemes do not provide reliable protection. This paper investigates different protection models based on machine learning approaches: support vector machine (SVM), decision tree (DT), and random forest (RF), which accurately detect the variability in the microgrid operation under normal and fault conditions. A MATLAB Simulink environment is employed to model a comprehensive microgrid system, incorporating PV systems, diesel generators, and diverse loads. Subsequently, RMS values of three-phase voltage and current measurements are seamlessly integrated into a Python codebase as data inputs to train and test the protection models. By employing simulation data, the models based on the three techniques (SVM, DT, and RF) learn to accurately identify the mode of operation, the variations in PV generation, and the occurrence of a fault and its type in the microgrid. The practicability of the proposed schemes is tested on a five-bus AC microgrid. Comparisons among approaches in terms of accuracy and speed of operation results are highlighted. The results show the improved accuracy and adaptability of the machine learning-based protection system compared to conventional methods.