| Resumo: | This thesis presents a methodology that applies a new automatic learning technique - the hybrid regression trees -, which exploits the functional knowledge about systems behavior. This methodology was specially conceived to exploit the dynamic behavior of isolated power systems with large wind power production. The hybrid regression trees, besides producing fast security classification of the system, can still quantify in real-time the security degree of the system by emulating continuos security indices that define the power system dynamic behavior. The application of the developed methodology to the power system of Terceira island is described. The main goal of this procedure was to obtain security structures, which produce fast dynamic security assessment of the system regarding frequency instability problems. A description of the data set generation procedure is included. This document also describes the application of the methodology to the power system of Crete. The main goal of this procedure was to develop advanced tools that can help operators to perform the management and operation of the installed wind power. The performance evaluation of the trained security structures, including comparative assessment with decision trees and neural networks, is also presented. The research leading to this thesis was developed within the framework of the last stage of the EU CARE project of the JOULE/THERMIE program. The work was carried out at FEUP (Faculdade de Engenharia da Universidade do Porto) and at INESC Porto (Instituto de Engenharia de Sistemas e Computadores do Porto).
|
|---|