Banca de DEFESA: DAVI GUIMARÃES DA SILVA

Uma banca de DEFESA de DOUTORADO foi cadastrada pelo programa.
STUDENT : DAVI GUIMARÃES DA SILVA
DATE: 03/04/2023
TIME: 14:00
LOCAL: Auditório do NTB / Google Meet - https://meet.google.com/wup-gwpz-zyk
TITLE:

Forecasting of Electric Energy Consumption Time Series with Deep Learning for an IoT System

KEY WORDS:

Deep Learning. Internet of things. Energy efficiency. Deep Neural
Networks. Prediction module.

PAGES: 138
BIG AREA: Outra
AREA: Ciências Ambientais
SUMMARY:

Energy consumption and energy efficiency are topics that have attracted the attention of researchers in recent years, in order to seek scientific and technological solutions for energy production and cost reduction. One of the alternatives that have obtained satisfactory results is the use of technologies based on Internet of Things (IoT) and Deep Learning (DL) systems. Based on this, we propose to evaluate the performance of Deep Neural Networks (DNN) to subsequently add it to a prediction module of an IoT-based system that performs real-time monitoring of electricity consumption. The work was divided into three parts as follows: 1) the performance of a deep neural network of the Long Short-Term Memory (LSTM) type was evaluated in relation to the number of layers with cross-validation of time series (CV-TS), on two datasets with electric power consumption data, where the results pointed out that there was no statistically significant difference between the layers; 2) CV-TS tests were performed on the same datasets used previously, to compare the results of the LSTM network with the Extreme Gradient Boost (XGBoost) and Random Forest (RF) algorithms, commonly used in the literature. The computational results and statistical tests (p < 0.0001 with Kruskal-Wallis test) indicated that the LSTM model showed a tendency to better Root Mean Square Error (RMSE) performance in both datasets; 3) finally, a new statistical comparison between the LSTM neural network and the Bidirectional Long Short-Term Memory (BiLSTM) neural network was performed, since it is a variation of the one-way LSTM used successfully in the previous tests, with validation data for two new datasets (in this case, all four datasets in total were used), with CV-TS, and in these tests, the BiLSTM network showed statistically significant better results in all datasets (p < 0.0001 with Friedman’s test). Thus, based on the results obtained where BiLSTM presented the best results in training, testing, and validation, it was decided to use it to compose the forecasting module of the IoT system, considering that from the results obtained by it in a real application, actions applicable to decision making can be implemented.

COMMITTEE MEMBERS:
Presidente - 1963026 - ANDERSON ALVARENGA DE MOURA MENESES
Externa à Instituição - ANDRESSA DOS SANTOS NICOLAU - UFRJ
Externa ao Programa - 2143011 - HELAINE CRISTINA MORAES FURTADO - nullExterno à Instituição - MARLA TERESINHA BARBOSA GELLER - ULBRA
Externo à Instituição - ROBERTO SCHIRRU - UFRJ