Advanced Feature Extraction Techniques for Machine Learning-Based Fault Analysis in Power Grids

As the infrastructure develops and more elements are added in the daily livelihood of humankind, dependable electrical power distribution is critical for keeping society and economies running smoothly. Modern power grids are complex, with wide networks and diverse components, requiring advanced problem detection and management systems for its upkeep. The growing demand of reliable power supply, as well as the necessity for long-term power infrastructure, emphasize the importance of reliable fault detection systems. This work aims to improve the resilience and reliability of electricity grids by using machine learning (ML) techniques to detect and analyze faults in power grids, using extensive feature extraction. The original base dataset, generated on an IEEE 6-bus grid system for consistent benchmarking, of three phase voltages and currents will be transformed into a dataset inclusive of time and other derivations, namely, voltage and current unbalance factor, volt-age and current harmonic distortion and the root-mean-square current and voltages. The results show a massive improvement in accuracy and related metrics between the base dataset, hovering between 34%-45% for various ML techniques to the improved dataset, yielding accuracy in the range of 87% -92% for respective techniques.