Aplicação de Modelos de Aprendizado de Máquina na Predição de Curvas de Carga
Urzagasti, Carlos Alejandro
MetadataShow full item record
One of the most important research topics in Smart Grid technology is load curve prediction. This allows electric power transmission and distribution operators to carry out operations with greater efficiency in the dispatch of electric power. Therefore, the relevance of knowing the characteristics of the electricity consumption profiles of consumers is highlighted, in order to then make load predictions, which can be done in different prediction horizons and with different methods. Therefore, this study aims to present machine learning techniques to perform load predictions within 30 minutes, ANN and LTSM techniques are analyzed. From a database containing load curve information of a 4-year distribution system (2013 to 2016), 4 machine learning models were made, two from RNA and two from LSTM. A comparison of both models was made, where a better performance was found for the LSTM models with accuracy greater than 0.95. It should be noted that this work only made a comparison for a single forecast horizon, therefore it remains as future work to analyze for different forecast horizons and search for new open databases to complement this study.