الفهرس 
INTRODUCTIONOnly 14 pages are availabe for public view
 |  | from | 74 |  |  |
| | | from | 74 | | |
Abstract
Forecasting in general is one of the hot topics helps in optimization and
management. Load forecasting in specific is very interesting topic that helps in
resources management and decision making especially in Smart Grid (SG).
SG is a modern electric grid with a two-way flow of electricity and data between
power utilities and consumers. Load forecasting helps in electricity management, such
that prior knowledge of needed electricity helped in the amount of electricity needed to
be generated.
Various models have been developed to forecast electrical load whatever for short
periods or long. Many of these models use their own techniques which are single
models. Other models are combined models that combine forecasts from more than one
model. Combined models either use linear techniques or nonlinear techniques for
combining forecasts from different models. Linear techniques are easy to understand
and implement. They use linear or equal weights for each of the contributing models
and totally ignore the relationships between the participating models and consider only
their contributions. So there is a considerable reduction of accuracy performance when
two or more of the participating models are correlated. Nonlinear techniques are
complex to understand and implement but they achieve better accuracy. Nonlinear
techniques consider contributions of the participating models as well as the
relationships among them, and that improves the performance accuracy when two or
more of the participating models are correlated.
In this thesis, a nonlinear weighted technique that combines forecasts of three of
the participating models is proposed. At the beginning, forecasts from four models
which are Random Forest (RF), Least Square Support Vector Machine (LSSVM),