| The use of new technological developments in electrical power systems monitoring together with advanced remote control devices and fast and reliable computer equipment offer the possibility of selecting the system configuration to reflect multiple attributes. Changes in the configuration of the system may take place to restore the system after a collapse by a fault, to improve the operation, reliability and robustness of the system, to minimize the environmental impact of the system or to optimize operational and other related costs, depending on the load requirements, equipment availability and condition of the system.; The scope of the research includes experiments that lead to the creation of an on-line, real time load flow based operations methodology for an electrical power system. This dissertation is intended to present an approach that will use a simple genetic algorithm as a teacher for the process of supervised learning of a feedforward, backpropagation artificial neural network for on-line control systems. The model is reached after several experiments that address-different considerations.; Initially, the use of historical data was considered, so that the decisions taken in the previous operation of the power system would become the rules to be followed by the artificial neural network. However, there was no warranty that the actions taken in the past would be the optimal ones, therefore, optimization techniques must be considered. Historical data was used to feed the genetic algorithm optimization program and this data along with the actions suggested by the optimization program would be used to train the artificial neural network. Instead of using the genetic programming in order to configure the power system a priori, it is used in order to configure and re-configure the system on-line.; This dissertation presents three experimental stages: the use of historical data of the conditions in the power system and the response that a human operator gave to each event, the use of a simple genetic algorithm in order to improve the response that would be given to each particular condition in the power system, and a combination of the first two (GA/ANN Approach). Two power engineering problems are considered in this research: a simplification of the automatic generation control problem and the optimal switching conditions problem. However the concepts that are used could be modified to be used in different decision making and optimization problems. Finally, data from a real power system is used as an example of application: the Mexican National Interconnected Power System, specifically the Merida Control Sub-Area of the Peninsular Control Area. |
