Detecting transient events with genetic algorithms

Abstract

Accurately detecting and analyzing events in a power system is a difficult, but important task. This project aims to determine whether genetic algorithms are viable - in terms of both accuracy and efficiency - for detecting these events in large sets of power systems data. Although power systems events are intially classified using a trigger-based system at the time of the event, this project aims to show that power systems can be broken down into their component parts (lineto-line voltage, current, etc.) and analyzed in isolation after the fact with similar accuracy. The system attempts to achieve this goal by iteratively forecasting the next value in a time series dataset and calculating the root-mean-squared error (RMSE) of the prediction, which is then averaged over the whole sample. This approach did yield substantive results - most importantly, this means that the two main assumptions that the project is based on were validated. The paper dives into the metrics generated by this approach in an effort to explain these results. The results of the experiment are discussed, and the paper is concluded by recommending future areas of development that could benefit the project.