-
Gabryel M. Raposo de Alencar
Electrical Engineering Department and Renewable Energy Engineering Department, Federal University of Paraíba, PB
-
Camila Seibel Gehrke
Electrical Engineering Department and Renewable Energy Engineering Department, Federal University of Paraíba, PB
-
Flávio da Silva V. Gomes
Electrical Engineering Department and Renewable Energy Engineering Department, Federal University of Paraíba, PB
-
Lucas Vinicius Hartmann
Electrical Engineering Department and Renewable Energy Engineering Department, Federal University of Paraíba, PB
-
Camila Maciel
Electrical Engineering Department and Renewable Energy Engineering Department, Federal University of Paraíba, PB
Keywords:
Time Series, Electric Power, Artificial Neural Networks, LSTM, CNN, Energy Optimization, EMS
Abstract
Energy management systems (EMS) has importance to utilities for several reasons, mainly for existing a necessity to allocate their resources in advance, requiring short, medium and long-term planning. Therefore, in this paper, a very short-term demand forecasting procedure was implemented, using a computational model based on Artificial Neural Networks (ANN) of the type Long Short-Term Memory (LSTM) to aid the Federal University of Paraíba (UFPB) analyzing peaks and off-peaks of active power throughout the past years being possible to use the forecast of One Hour Ahead as a input to a BESS for optimal battery power dispatch management, in order to reduce fines for excess contracted demand at UFPB. The study compares LSTM with a CNN model and to improve the performance of the LSTM, the network was evaluated under different aspects, as hyperparameters and considered a technique of periodicity using sines and cosines.
