[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT REQUEST OF AUTHOR.] Power distribution systems are designed and constructed as closed loops, but they are operated in a radial topology by choosing suitable open tie switches. Radial configurations are used because they satisfy various operational and protection requirements. Distribution system reconfiguration (DSR) determines the status of both sectionalizing switches (which are normally closed) and tie switches (which are normally open). DSR has great benefits in both normal and abnormal operations (outages). DSR is a multi-objective, non-linear problem. A new, fast, non-dominated sorting genetic algorithm (FNSGA) is introduced for solving the DSR problem in normal operation by satisfying all objectives simultaneously with a relatively small numbers of population size and generations and short computational time. The dissertation describes creative contributions to genetic algorithm science for the DSR problem and describes results of applying the FNSGA to a standard IEEE test system. The results show the efficiency of this algorithm as compared to other methods in terms of both achieving all the goals and minimizing the computational time with reasonable population and generation sizes. The objectives of the problem in normal operation are to optimize the system performance and efficiency in terms of maximizing the operating voltage and minimizing the branch loading. The operation cost will be reduced by minimizing the real power losses. This should be achieved with a small number of switching operations. The objectives of the problem in normal operation are to minimize real power losses and improve the voltage profile and load-balancing index with minimum switching operations. In this dissertation, a load shedding strategy based on priority customers, minimization of the number of affected buses, and minimization of the number of switching operations is introduced. To test the algorithm, it was applied to three widely studied test systems and a real one. The results show the efficiency of this algorithm as compared to other methods in terms of achieving all the objectives simultaneously with reasonable population and generation sizes and without using a mutation rate, which is usually problem-dependent.
