الفهرس 
Description of Electric Power System and Smart GridsOnly 14 pages are availabe for public view
 |  | from | 122 |  |  |
| | | from | 122 | | |
Abstract
The complexity of decision making in power electric market due to the
deregulation that happens in economy and energy generation and purchasing,
demands different levels of decision to be taken in consideration. When applying that in the load forecasting problem, it leads to accurate and reliable information about power system. It also causes a very good management to economics, reliable
and stable operation of the power system. To face such complexities, load demand prediction become required a strong accuracy and no more errors in its values, where it depends on the form of planning and the accuracy required. For all that reasons, electric load forecasting has more and more importance in the electrical
sector.
Classical methods of the electrical load forecasting problem, lack to grasp accurately the non-linearity that happens in the load curve and also the random changes that occur in the electrical load affecting features. Because of all of these
drawbacks in the classical methods, the most recent trends to solve the candidate problem are artificial intelligence. Meta-heuristic optimization algorithms, which
are considered from artificial intelligence techniques, are designed to find good solutions to optimization problems with incomplete or limited computation capacity. They have a huge ability for unscrew the complex problems node, for example, genetic algorithm (GA), ant colony algorithm (ACO), particle swarm algorithm (PSO), firefly algorithm (FFO), fruit fly algorithm (FOA) and others which deal with multi-objective complex optimization problems, where the
electrical load forecasting problem is considered. Artificial neural networks (NN) was applied to the load forecasting problem and it includes feed forward neural network (FFN) which is a type of multi-layer perceptron (MLP), radial base function network (RBF), recurrent neural network (RNN), and spiking neural
networks which improve the accuracy of electrical load forecasting process.
The focus in this study is on the long term load forecasting process (LTLF) to add a contribution in overcoming the problem that face researchers and
participants in this field.
This thesis introduces new evolutionary hybrid models using the grey wolf optimization algorithm and its modified versions to train FFN to perform LTLF process, these models are validated by comparing with the hybrid model of PSO and FFN and other types of NN, where the proposed models proves their superiority and ability to make the forecasting process more accurate and reliable.
This thesis consists of six main chapters. These chapters can be described in the following manner:
CHAPTER 1: The most important aim of this chapter is to introduce the basic
concepts and definitions of the electrical load forecasting process, its classifications, classical and modern methods of forecasting and nonlinear programming optimization and its classifications, and then introducing some of improved works in electrical load forecasting process.
CHAPTER 2: This chapter introduces an improved regression analysis by
hybridizing it with a proposed grey wolf optimization algorithm (GWO). In order to validate the proposed algorithm a comparison is implemented between it and PSO when hybridizing with regression methods. The proposed models are applied to forecast annual electrical peak load of Jordan.
CHAPTER 3: This chapter proposes a modified grey wolf optimization (MGWO)
and hybridizing it with feed forward neural network (FFN) to improve its training process, namely (MGWO-FFN). The proposed MGWO-FFN is utilized to
implement annual electrical load forecasting of Egypt and is validated by comparing it with the hybrid models of PSO-FFN and GWO-FFN. The work is tested by comparing with another work of forecasting annual peak load of
Beijing’s city, where the proposed MGWO-FFN outperforms the other models
used.
CHAPTER 4: This chapter proposes another modification of grey wolf optimizer and utilizes it to train FFN namely MGWO-FFN which is utilized to forecast the annual peak load of Jordan through two different strategies of LTLF. The first
strategy is to predict the load when utilizing more than one feature like (population growth, gross domestic product and the historical annual electrical peak load). But the second strategy is to predict the load using only one feature (historical annual
peak load).
CHAPTER 5: This chapter proposes a robust solution to LTLF problems which is summarized into two parts, one is to cluster the raw data of annual electrical load linearly to address the fluctuations in some activities for which electricity is distributed, and the second part is to eliminate the error caused by the model by
implementing a hybrid synergy model that merges two hybrid models of PSO and GWO each to train feed forward neural network which give more accurate results. Work is compared with several types of neural networks. The first study case that is introduced is to forecast annual electrical load consumed of Egypt, by gathering data of each activity for which electricity is distributed and then these raw data is linearly clustered and prepared for forecasting implementation. The second case is
to forecast the shape of the future energy sources in Egypt by gathering the historical electrical power generated for several sources which are new and renewable energy, natural gas energy, hybrid fuel energy, hydraulic fuel energy and steam energy, then the raw data is preprocessed also and forecasted, finally the total annual electrical power generated is gained by summing results of each preprocessed set. A percentage evaluation is estimated to identify the contribution of each source in the total electrical power generated.
CHAPTER 6: This chapter describes some concluding remarks, recommendations, and some points for further researches.