Deep Learning Models for Stability Monitoring of Smart Grids

Abstract

Increased interest in renewable sources as a result of significant energy and environmental issues, as well as the development of dedicated technologies, has led to the conception of a smart grid. However, the implementation of smart electrical grids presents important challenges
that require attention to ensure proper operation in all aspects and prevent severe blackouts. One of those challenges is monitoring the conditions of stability in the network, both to guarantee its integrity and to make it economically feasible. In the case of decentralized
networks, stability instantly depends on electricity prices, availability and reaction times by consumers and generators. In the field of smart grids, systems based on deep learning techniques have proven their considerable performance in a variety of tasks. In this work, deep learning models are proposed with the main objective of predicting unusual events of instability in a decentralized intelligent network. Later, we employed a missing data imputation to simulate interruptions in the collection of process data and evaluated how data imputation affected deep learning models. Results show the improved performance of the implemented deep learning algorithms under different operating conditions.