| Abstract: |
Smart electrical grids are a new paradigm in the power distribution systems; they combine the new technologies in communication and monitoring to increase its efficiency and reliability. The current paper examines machine learning as an intelligence tool to detect faults in smart electric grids and locate them. The main aim is to compare the performance of different machine learning algorithms such as Support Vector machine, Random Forest, Convolutional Neural Network and Long Short-Term Memory networks based on proper fault detection, classification and location estimation. It used a quantitative research methodology that involved the use of secondary data on IEEE standard bus systems and published research datasets. The hypothesis formulated that deep learning solutions were more accurate than in comparison with traditional machine learning solutions in fault diagnosis applications. It has been shown that hybrid CNN-LSTM models offer classification rates of above 99 percent, whereas the less complex algorithm of Random Forest with the same accuracy of 97-98 percent. It can be seen in discussion that the occurrence of single-line-to-ground faults is around 70-80 percent of the total grid faults, which is critical in justifying the importance of powerful detection systems. The research finds that the implementation of machine learning in smart grid protection systems will greatly improve the fault management capabilities, but the issues of data quality, cybersecurity, and real-time implementation need further investigation. |
