Feature Selection and Optimization of a ML Fault Location Algorithm for Low Voltage Grids

Low voltage (LV) grids are changing drastically, entering a new era of digitalization, broad renewable energy integration and active consumer participation in the energy management. The challenges and opportunities created by these changes affect also the protection of the grids. The already limited fault location methods developed for LV grids are quickly becoming outdated. At the same time the growing volume of available data facilitates the employment of the powerful, rapidly advancing artificial intelligence tools. Here, a feature selection technique was combined with a random forest in order to ensure the minimization of the required data and the optimization of the process. The optimized model achieved an test set accuracy of 99%. The training and verification of the algorithm were performed on a modified version of the CIGRE European LV benchmark. The research was complemented by a sensitivity analysis on the effect of the fault resistance and the photovoltaic power generation on the algorithm’s accuracy.