Application of PSO for Optimization of Power Systems under Uncertainty

Doctoral Thesis / Dissertation from the year 2009 in the subject Electrotechnology, grade: 1.0, University of Duisburg-Essen (Institute of Electrical Power Systems), course: Electrical Engineering, language: English, abstract: The primary objective of this dissertation is to develop a black box optimization tool. The algorithm should be able to solve complex nonlinear, multimodal, discontinuous and mixed-integer power system optimization problems without any model reduction. Although there are many computational intelligence (CI) based algorithms which can handle these problems, they require intense human intervention in the form of parameter tuning, selection of a suitable algorithm for a given problem etc. The idea here is to develop an algorithm that works relatively well on a variety of problems with minimum human effort. The most significant optimization task in the power system operation is the scheduling of various generation resources (Unit Commitment, UC). The current practice used in UC modelling is the binary approach. This modelling results in a high dimension problem. This in turn leads to increased computational effort and decreased efficiency of the algorithm. A duty cycle based modelling proposed in this thesis results in 80 percent reduction in the problem dimension. The stern uptime and downtime requirements are also included in the modelling. Therefore, the search process mostly starts in a feasible solution space. From the investigations on a benchmark problem, it was found that the new modelling results in high quality solutions along with improved convergence. The final focus of this thesis is to investigate the impact of unpredictable nature of demand and renewable generation on the power system operation. These quantities should be treated as a stochastic processes evolving over time. A new PSO based uncertainty modelling technique is used to abolish the restrictions imposed by the conventional modelling algorithms. The stochastic models ar