Ewaldo Santana

Extraction of Signals With Specific Temporal Structure Using Kernel Methods

This work derives and evaluates a method for Blind Source Extraction (BSE) in a reproducing kernel Hilbert space (RKHS) framework. The a priori information about the autocorrelation function of the signal under study is translated in a linear transformation of the Gram matrix of the transformed data in Hilbert space. Our method proved to be more robust than methods presented in the literature of BSE with respect to ambiguities in the available a priori information of the signal to be extracted. The approach here introduced can also be seen as a generalization of Kernel principal component analysis (KPCA) to analyze autocorrelation matrices at specific time lags.

An adaptive recursive algorithm based on non-quadratic function of the error

In adaptive filtering, several algorithms were developed to get faster convergence and lower misadjustment, but rely on second order statistics which are optimum only for Gaussian signals. In this work we propose a recursive filter by modifying the performance surface to a non-quadratic function applied upon the error. As a result, the equations are simple, elegant, and yielded faster convergence and lower misadjustment when compared to the RLS, keeping equivalent computational cost.

Diabetes classification using a redundancy reduction preprocessor

Introduction Diabetes patients can benefit significantly from early diagnosis. Thus, accurate automated screening is becoming increasingly important due to the wide spread of that disease. Previous studies in automated screening have found a maximum accuracy of 92.6%. Methods This work proposes a classification methodology based on efficient coding of the input data, which is carried out by decreasing input data redundancy using well-known ICA algorithms, such as FastICA, JADE and INFOMAX. The classifier used in the task to discriminate diabetics from non-diaibetics is the one class support vector machine. Classification tests were performed using noninvasive and invasive indicators. Results The results suggest that redundancy reduction increases one-class support vector machine performance when discriminating between diabetics and nondiabetics up to an accuracy of 98.47% while using all indicators. By using only noninvasive indicators, an accuracy of 98.28% was obtained. Conclusion The ICA feature extraction improves the performance of the classifier in the data set because it reduces the statistical dependence of the collected data, which increases the ability of the classifier to find accurate class boundaries.

Estimators Based on Non-squares Loss Functions to Approximate HJB-Riccati Equation Solution for DLQR Design via HDP

This paper is concerned with the development of online algorithms for approximate solutions of the Hamilton-Jacobi-Bellman (HJB) equation. In the discrete linear quadratic regulator (DLQR) control system design, the HJB equation is the discrete algebraic Riccati (DARE) equation. Due to the problem of dimensionality curse, this equation is approximated via heuristic dynamic programming (HDP). The proposed methodology is based on a familiy of non-squares approximators for critic adaptive solution of the DARE associated to the DLQR problem, referred to in this work as HJB-Riccati equation, which is characterized as a parameterization of the HJB equation. The proposed method is evaluated in a multivariable dynamic system of 4th order with two inputs and it is compared with standard recursive least square algorithm.

An algorithm based on non-squared sum of the errors

In adaptive filtering, several algorithms are developed in the quest for greater convergence speed, mostly relying on second order statistics. Here we modify the Recursive Least Square (RLS) equations by using as performance surface a weighted sum of even error power. As a result, the equations turn out to be simple, elegant, while yielding faster convergence and preserving the computational cost when compared with the existing RLS algorithm. HighlightsIn this study, we propose a new recursive algorithm that optimizes a sum of the even powers of the error.In adaptive filtering, several algorithms were developed to improve convergence speed, but they rely mostly on second order statistics.This algorithm seems to be faster than usual recursive algorithms presents in the literature.

Convergence Analysis using non-squares estimators to approximate the solution of HJB-Riccati equation for the design DLQR via HDP

The proposed methodology is based on development of online algorithms for approximate solutions of the Hamilton-Jacobi-Bellman (HJB) equation through a family of non-squares approximators for critic adaptive solution of the Discrete Algebraic Riccati Equation (DARE), associated with the problem of Discrete Linear Quadratic Regulator (DLQR). The proposed method is evaluated in a multivariable dynamic system of 4th order with two inputs and it is compared with standard recursive least square algorithm.

Optimal control of a wind generator system using non-squares estimators

The control of eolic and solar energy systems demands methods and technics adapted to the high degree of environment non-stationarities whose adjustments are carried out via adaptive filters. Among the best known are least mean square (LMS) and the recursive least square (RLS) algorithms [1] and [2]. However, those algorithms still fail to respond quickly to the optimal control of the doubly fed induction generator (DFIG) as required in online learning [3]. Here we propose a methodology based on approximate solutions to the linear quadratic regulator (LQR) by using a family of non-squares approximations [4], [5]. We show experimentally that the RLNS provides more accurate estimates for DLQR when compared to the RLS while showing a convergence speed to the actual solution in less than 50% of the iterations as required by the standard RLS estimator for approximating Ricatti equation solution via Heuristic Dynamic Programming (HDP) [6].

RLS Algorithms and Convergence Analysis Method for Online DLQR Control Design via Heuristic Dynamic Programming

In this paper, a method to design online optimal policies that encompasses Hamilton-Jacobi-Bellman (HJB) equation solution approximation and heuristic dynamic programming (HDP) approach is proposed. Recursive least squares (RLS) algorithms are developed to approximate the HJB equation solution that is supported by a sequence of greedy policies. The proposal investigates the convergence properties of a family of RLS algorithms and its numerical complexity in the context of reinforcement learning and optimal control. The algorithms are computationally evaluated in an electric circuit model that represents an MIMO dynamic system. The results presented herein emphasize the convergence behaviour of the RLS, projection and Kaczmarz algorithms that are developed for online applications.

Measuring the excitation current in transformers using hall effect sensors

This paper presents and discusses the use of Hall effect sensors to measure the excitation current in single-phase transformers. The Hall effect sensors have galvanic isolation of the order of a few kVrms and may have total measurement errors less than 0.7% of the nominal value. It is common in transformer tests, to use a varivolt to avoid inrush current, which can reach more than 20 times the rated operating current, and can interfere with or even damage electronic circuits for measurement. However with the use of these sensors, their characteristics make it possible to have a measuring device relatively robust, accurate and fast, thus eliminating the use of a varivolt, for testing the load current transformer single phase or three phase.

Extraction of signals with higher order temporal structure using Correntropy

This paper addresses the problem of semi-blindly extracting one single desired signal using a priori information about its higher order temporal structure. Our approach is based on the maximization of the autocorrentropy function for a given time delay. The a priori information is quantified as a time delay and a size for a Gaussian kernel to set the free parameters in the correntropy function. Those values provide information which allows the proposed method to adapt a demixing vector to extract the desired signal without the indeterminacy of the permutation problem in blind source separation. Moreover, this method is different from those for Independent Component Analysis that separate all the available sources, which, in some problems, is not desirable or computationally possible. Since correntropy can be interpreted as a generalization of correlation, we demonstrate that it is a suitable measure for studying the temporal behavior of higher order statistics of a signal. Also, the flexibility brought by the kernel size selection allows the user to choose the range of statistics he is interested in. We show in simulations that correntropy achieve better or equal separation than other linear methods proposed in the literature for source extraction based on temporal structures.