Fault Classification in Networks with Renewable Energy Sources Using Machine Learning

Fault classification is the first step to mitigation in power system network, and the unique fault current profile of inverter interfaced renewable energy source causes malfunction of classifiers that works on conventional system. Inverters have internal protection schemes which trip during unbalance conditions. However, for high penetration of renewable energy, the inverter must ride through the fault and let relays to protect the system. This study proposes a novel machine learning based fault classification methodology that works on raw values of phase voltage in the presence of an inverter interfaced photovoltaic or wind turbine system. Incremental value and rate of change of phase voltage magnitude forms the basis of features used in a random forest algorithm. Measurements only from the local end of line are required to identify the fault. The performance of the classifier is validated against different fault scenarios in systems like IEEE 9 bus and 13 bus distribution feeders. The proposed classifier achieved a satisfactory accuracy for test conditions within three quarters of a cycle and can be used for any three-phase system to provide correct faulted phase information to other protection components. The same methodology is extended to identify evolving faults with 99.6% accuracy