Data-Driven Indicators Applied to Power Distribution Systems

Indicators play an important role as they offer a quick overview of the system performance. However, traditional indicators were conceptualized as a static indicator that was computed on a specific time, making it difficult to extrapolate these indicators to the future. This task is more challenging in large scale systems as in power distribution systems, due to the stochastic behavior of time-varying loads such as electric vehicles (EV), distributed energy resources (DERs), and conditions of reconfiguration due to microgrids presence. In this work, we explain in detail how to transform any indicator in a data-driven indicator. The main objective of data-driven indicators is to aid decision-makers to improve the understanding of the behavior of indicators along the time, even in the near future. Our proposed methodology, use a long short-term memory (LSTM) auto-encoder architecture as a feature extractor in order to find a latent space of our historic indicators, then we used an LSTM forecaster network to forecast our indicators, from the latent space with a unique model. We carried out this analysis using real-world data collected at the University of Campinas (UNICAMP) obtained from 300 smart meters. We obtained an accuracy RMSE = 1.29 in our forecasted indicators.