Eliana Mirledy Toro

IDENTIFICACIÓN EFICIENTE DE ERRORES EN ESTIMACIÓN DE ESTADO USANDO UN ALGORITMO GENÉTICO ESPECIALIZADO (EFFICIENT IDENTIFICATION OF ERRORS IN STATE ESTIMATION THROUGH A SPECIALIZED GENETIC ALGORITHM)

En este articulo se presenta un metodo para resolver el problema de estimacion de estado en sistemas electricos usando optimizacion combinatoria. Su objetivo es el estudio de mediciones con errores de dificil deteccion, que afectan el desempeno y calidad de los resultados cuando se emplea un estimador de estado clasico. Dada su complejidad matematica, se deducen indicadores de sensibilidad de la teoria de puntos de apalancamiento que se usan en el algoritmo de optimizacion de Chu-Beasley, con el fin de disminuir el esfuerzo computacional y mejorar la calidad de los resultados. El metodo propuesto se valida en un sistema IEEE de 30 nodos Abstract: In this paper a method to solve the state estimation problem in electric systems applying combinatorial optimization is presented. Its objective is the study of measures with difficult detection errors, which affect the performance and quality of the results when a classic state estimator is used. Due to the mathematical complexity, sensibility indicators are deduced from the theory of leverage points used in the Chu-Beasley optimization algorithm with the purpose of reducing the computational effort and enhance the quality of the results. The proposed method is validated in a 30-node IEEE system.

IDENTIFICACIÓN DE ERRORES GRANDES EN LA ESTIMACIÓN DE ESTADO USANDO BÚSQUEDA TABÚ

The state estimation for power systems is commonly developed using the weighted-least-square (WLS) technique. Nevertheless, difficulties in the identification of erroneous measurements appear, since there isn’t a verification process that allows to have certainty on the quality of these measurements. Therefore, in this article is shown a methodology that combines the classic technique with a method of combinatorial optimization to guarantee a greater efficiency in the detection of erroneous measurements. In addition, the theory of leverage points is used like observability criterion.

Optimización matemática en superficies de desempeño de redes neuronales artificiales

Several optimization methods are used in this article in order to show the behavior on performance surfaces and to conclude about efficiency in neuronal networks training. These methods are: descendent gradient, Newton Raphson, conjugated gradient and its adaptations to multilayer networks: basic backpropagation and Levenberg-Marquardt.