Carlos D. Zuluaga

A Parzen-Based Distance Between Probability Measures as an Alternative of Summary Statistics in Approximate Bayesian Computation

Approximate Bayesian Computation (ABC) are likelihood-free Monte Carlo methods. ABC methods use a comparison between simulated data, using different parameters drawn from a prior distribution, and observed data. This comparison process is based on computing a distance between the summary statistics from the simulated data and the observed data. For complex models, it is usually difficult to define a methodology for choosing or constructing the summary statistics. Recently, a nonparametric ABC has been proposed, that uses a dissimilarity measure between discrete distributions based on empirical kernel embeddings as an alternative for summary statistics. The nonparametric ABC outperforms other methods including ABC, kernel ABC or synthetic likelihood ABC. However, it assumes that the probability distributions are discrete, and it is not robust when dealing with few observations. In this paper, we propose to apply kernel embeddings using a sufficiently smooth density estimator or Parzen estimator for comparing the empirical data distributions, and computing the ABC posterior. Synthetic data and real data were used to test the Bayesian inference of our method. We compare our method with respect to state-of-the-art methods, and demonstrate that our method is a robust estimator of the posterior distribution in terms of the number of observations.

Approximate Probabilistic Power Flow

Power flow analysis is a necessary tool for operating and planning Power systems. This tool uses a deterministic approach for obtaining the steady state of the system for a specified set of power generation, loads, and network conditions. However this deterministic methodology does not take into account the uncertainty in the power systems, for example the variability in power generation, variation in the demand, changes in network configuration. The probabilistic power flow (PPF) study has been used as an useful tool to consider the system uncertainties in power systems. In this paper, we propose another alternative for solving the PPF problem. This paper shows a formulation of the PPF problem under a Bayesian inference perspective and also presents an approximate Bayesian inference method as a suitable solution of a PPF problem. The proposed method assumes a solution drew from a prior distribution, then it obtains simulated data (active and reactive power injected) using power flow equations and finally compares the observed data and simulated data for accepting the solution or rejecting these variables. This overall procedure is known as Approximate Bayesian Computation (ABC). An experimental comparison between the proposed methodology and traditional Monte Carlo simulation is also shown. The proposed methods have been applied on a 6 bus test system and 39 bus test system modified to include a wind farm. Results show that the proposed methodology based on ABC is another alternative for solving the probabilistic power flow problem; similarly this approximate method take less computation time for obtaining the probabilistic solution with respect to the state-of-the-art methodologies.

Identificación robusta aplicada a un sistema de control de un puente grúa

This study presents a methodology to perform the robust identification of a gantry crane control system. The robust modeling was performed by three sequential Kalman filters, where two of them are the dual Kalman filter for estimating the parameter and state system, and third filter is the robust statistic Kalman filter. The obtained results show that the robust statistic Kalman filter presents responses smoother for the position and the velocity of the gantry crane, compared with the standard identification strategies.

Regularización de problemas dinámicos inversos en la generación EEG mediante estimación dual basada en el filtro de Kalman

This study presents the applications of two sequential Kalman filters to perform dynamic inverse problems regularization as the reconstruction of current distributions in neural activity in the brain, from electroencephalography signals. Kalman filter is an efficient algorithm for reconstructing of optimal way the current densities under some operation hypothesis, these are: the relationship between consecutives state; among a state and an observations, are both given by Gaussian distributions. The proposed methodology obtains consistent results with the state-of-the-art, when sources numbers rise; however, it needs a change in the estimation structure, since it can incur high computational cost.

Identificación de un generador de inducción doblemente alimentado basado en el filtro de Kalman en presencia de datos espurios

Este documento presenta una metodologia de identificacion de un generador de induccion doblemente alimentado(DFIG) en presencia de datos espurios. El DFIG es ampliamente usado en la produccion de energia electrica a partirdel viento; un problema en el control de estas maquinas, es el cambio en los parametros del sistema, haciendo queel esquema de control no tenga un optimo desempeno. Ademas, si el sistema sensorial no es confiable, se puedeincurrir en que las mediciones contengan datos espurios. Para llevar a cabo la identificacion se emplea tres filtros deKalman secuenciales, dos de ellos corresponden al filtro de Kalman dual, el otro corresponde al filtro de Kalman deestadistica robusta. La metodologia se implemento en Matlab, mostrando que la tecnica no se ve afectada por datosespurios, obteniendo errores residuales menores al 1.2% para la identificacion del DFIG en presencia de estos datos.

Tecnicas de seguimiento de caracteristicas faciales en secuencias de imágenes basadas en metodos de libre modelo.

Facial features play a important role in developmen t of systems on computer vision for different applications such as: the huma n-computer interactive, facial expressions automatic recognition also identify fat igue in car drivers, within these systems, facial features tracking is a signif icant stages that need to be developed, Therefore, this paper focuses on review the different techniques for facial features tracking, such as model-free method s (Kalman filter, particle Filter, among other).