Banca de DEFESA: DIEGO BARBOSA ARIZE SANTOS

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
DISCENTE : DIEGO BARBOSA ARIZE SANTOS
DATA : 14/10/2019
HORA: 14:00
LOCAL: SALA 12 - IME
TÍTULO:

Towards an accurate energy forecast given uncertainties in Smart Grids

PALAVRAS-CHAVES:

SMART GRIDS, TIME SERIES, FUZZY SYSTEMS

PÁGINAS: 75
GRANDE ÁREA: Ciências Exatas e da Terra
ÁREA: Ciência da Computação
SUBÁREA: Sistemas de Computação
ESPECIALIDADE: Arquitetura de Sistemas de Computação
RESUMO:

Managing energy generation and its consumption is usually a difficult task for electric power grids systems, given the great volume of data. Moreover, the data obtained from these systems is tied to uncertainty occasioned by seasonality and natural environment disturbances. Therefore, efforts have been made on the construction of Smart Grids, i.e. intelligent energy networks, which combine Computational Intelligence with the electricity power grids, in order to improve the balance between energy generation and its consumption. Smart Grids require powerful controllers to keep the balance of energy generation and its demand. Those controllers have to be aware of future loads, and the prediction of this data must be very accurate to provide efficient decision support. Since Smart Grid's energy generation and consumption data varies over time, following a time series distribution, time series forecasting methods can yield predictions to support those controllers on the decision-making process. Nevertheless, forecasting over data that tied to uncertainty may have some disturbances. In order to overcome those issues, on this work, we investigate time series forecasting methods, mapped in a systematic literature review, aiming to deliver accurate forecasts even for uncertain data. Towards finding a method for that, we present a comparison study over different time series forecasting methods to evaluate which one would achieve better accuracy in energy distribution. The compared methods were the Adaptive Fuzzy Neural Network (ANFIS), Recurrent Neural Networks (RNN), Support Vector Regression (SVR), Random Forest and SARIMAX. The ANFIS algorithm had outperformed the other approaches, delivering more accurate results. From that comparison, we have proposed a framework that combines the ANFIS' fuzzification step with well-known forecasting methods to improve their performances on forecasting under uncertain energy data.

MEMBROS DA BANCA:
Presidente - 2115505 - TATIANE NOGUEIRA RIOS
Interno - 2810986 - MARCOS ENNES BARRETO
Externo à Instituição - MATHEUS GIOVANNI PIRES - UEFS