الفهرس 
AbstractOnly 14 pages are availabe for public view
 |  | from | 106 |  |  |
| | | from | 106 | | |
Abstract
Coordination and control of power generation sources, both conventional and renewable, is one of the essential tasks of modern smart grids. Power generation is controlled through unit commitment (UC) and economic dispatch (ED) methods. With the increased penetration of power generation using renewable sources such as wind and solar, the complexity of the UC/ED problems increases due to the stochastic nature of renewable power. Therefore, an accurate and fast optimization method is needed when the generation process involves a large penetration of renewable sources to effectively manage the generation mix and load requirements. In this thesis, a forecasting model is built to simulate the stochastic nature of the renewable sources. A hybrid Markov method is proposed to forecast solar radiation, and an improved autoregressive integrated moving average (ARIMA) model is proposed to predict wind speed. ARIMA is improved by taking into account the non-stationary characteristic of the wind speed with few model parameters. The benefit of Markov theory is augmented by integrating Markov forecasting with Markov analysis, and incorporating a weather matrix with measured data to produce accurate forecasting data. This thesis presents a comparison between the proposed forecasting methods and the conventional methods, proving the superiority of the proposed methods in terms of accuracy and ease of application. For high penetration of renewable energy, a storage media must be used to compensate the fluctuation of the natural sources. In this thesis, sodium-sulfur (NaS) batteries are used as storage units. The main advantage of this type is long life, fast response and long discharge cycles. Also, this thesis introduces a model representing a charging and discharging process for storage units. Given the expensive price of storage units, a mathematical formulation is created to decide the best size of the battery in order to achieve the greatest benefit. This thesis also discusses UC analysis. The objective function of UC aims to minimize the total operational cost, which includes fuel cost, environmental cost, operation and maintenance cost,startup cost and shutdown cost. This objective function is subjected to a number of constraints; the system constraint, thermal units constraints, renewable sources constraints, and storage units constraint. The reserve constraint is modified to be a dynamic reserve to overcome the impact of the intermittent behavior of renewable sources and forecasting error, by inserting two reserve coefficients, up spinning reserve and down spinning reserve. UC is solved by using dynamic programming, simulated annealing, interior point, genetic algorithms, and particle swarm optimization application. The thesis introduces a mix of optimization techniques, where dynamic programming based on neural network is used to solve UC problem. Finally, a model predictive control (MPC) is introduced to solve the power system ED problems with the presence of renewable energy resources. Because the ED problem is a nonlinear, non-convex and mixed integer problem, the MPC is used in a nonlinear form. To produce a nonlinear model predictive control (NMPC), MPC must be integrated with a fast optimization methodology. The thesis presents a novel mathematical formula for a NMPC, integrated with swarm optimization technique to describe the nonlinear behavior of the problem. This new formulation is called swarm model predictive control (SMPC) optimization. The control model will be able to address the effect of the system disturbances and fluctuations, using a controlled autoregressive integrated moving average (CARIMA). The more accurate prediction leads to better controlled and optimized performance. A general form of future predictions can be expressed as a function of input and output past data, and a future control sequence, and is actually the degree of freedom in the SMPC problem. Also, the prediction part improves the swarm technique, because it better identifies the size of search space. In this thesis, UC schedule is designed using the swarm technique, while ED is solved using the proposed swarm model predictive dispatch (SMPD) optimization method. UC schedule is operated offline, while the dispatch decisions are computed in real time control using SMPD and fed into the automatic generation control (AGC) system.