As obras disponibilizadas nesta Biblioteca Digital foram publicadas sob expressa autorização dos respectivos autores, em conformidade com a Lei 9610/98.
A consulta aos textos, permitida por seus respectivos autores, é livre, bem como a impressão de trechos ou de um exemplar completo exclusivamente para uso próprio. Não são permitidas a impressão e a reprodução de obras completas com qualquer outra finalidade que não o uso próprio de quem imprime.
A reprodução de pequenos trechos, na forma de citações em trabalhos de terceiros que não o próprio autor do texto consultado,é permitida, na medida justificada para a compreeensão da citação e mediante a informação, junto à citação, do nome do autor do texto original, bem como da fonte da pesquisa.
A violação de direitos autorais é passível de sanções civis e penais.
This dissertation investigates the application of
Artificial Neural Networks (ANNs) in load forecasting. In
this work we have used real load data from the Brazilian
electrical system. The dissertation is divided in four
main topics: a study of the importance of load forecasting
to electric power systems; the investigation of the ANN
modeling to this particular problem; the development of a
neuro-simulador; and the case studies.
It has been made an investigation of the objectives of
load forecasting to power systems. The different kinds of
load forecasting have been classified according to the
leading time of the prediction (short and long term). The
more important variables to model electric load were also
investigated. This study analyses many projects in the
area of load forecasting and presents the techniques that
have been traditionally used to treat the problem.
The ANNs modeling to load forecasting involved a deep
investigation of works that have been published. The ANNs
architectures and learning algorithms more commonly used
were studied. It has been verified that the
Backpropagation algorithm was the more commoly applied in
the problem (particularly, in the problem of short term
hourly load forecasting). Based on this investigation and
using the backpropagation algorithm, many Neural Networks
architetures were proposed according to the desired type
The development of the neuro-simulator has been made in C
language, using SUN workstations. The software is divided
in 3 modules: a load series pre-processing module, to
prepare the input data; a training module to the load
series behavior learning; and an execution module, in
which the Neural Network will perform the predictions. The
development of a friendly interface to the forecasting
system execution and the portability of the system were
main goals during the simulator development.
The case studies involved testing the system performance
for 2 cases: hourly and monthly predictions. In the first
case, load data from CEMING (State of Minas Gerais) and
LIGHT (State of Rio de Janeiro) has been used. In the
second case load data from 32 companies of the Brazilian
electrical system has been used. Monthly load forecasting
is involved in a project of interest of two companies of
the electric sector in Brazil: CEPEL and ELETROBRÁS. In
both cases, influences of the forecasting horizon and of
the period of the year in the system´s performance has
been investigated. Besides, the changes in the forecasting
performance according to the particular electric company
were also studied. The performance evaluation has been
done through the analysis of the following error figures:
MAPE (Mean Absolute Percentage Error), RMSE (Root Mean
Square Error) and Theil´s U. The ANN performance was also
compared with the performance of other techniques, like
Holt-Winteres and Box-Jenkins, giving better results in