Catálogo de publicaciones - libros

This paper investigates modeling of the pulsed gas tungsten arc welding (GTAW) process using support vector machine (SVM). Modeling is one of the key techniques in the control of the arc welding process, but is still a very difficult problem because the process is multivariable, time-delay and nonlinear. We analyze the characteristics of SVM for solving the challenge problem and give the main steps of modeling, including selecting input/output variables, kernel function and parameters according to our specific problem. Experimental results of the SVM, neural network and rough set methods show the feasibility and superiority of our approach.

This paper presents a simple neural Automatic Voltage Regulator (AVR) for power systems. By representing the proposed neuro-controller in -domain, its parameters can be obtained analytically to ensure system stability. Results of simulation studies on a nonlinear third order generator model with the proposed neuro-controller using calculated parameters are given. Results of simulation studies demonstrate the effectiveness of this simple neuro-controller.

Estimating the correct location of electric current source with the brain from electroencephalographic (EEG) recordings is a challenging analytic and computational problem. Specifically, there is no unique solution and solutions do not depend continuously on the data. This is an inverse problem from EEG to dipole source. In this paper we consider a method combining backpropagation neural network (BPNN) with nonlinear least square (NLS) method for source localization. For inverse problem, the BP neural network and the NLS method has its own advantage and disadvantage, so we use the BPNN to supply the initial value to the NLS method and then get the final result, here we select the Powell algorithm to do the NLS calculating. All these work are for the fast and accurate dipole source localization. The main purpose of using this combined method is to localize two dipole sources when they are locating at the same region of the brain. The following investigations are presented to show that this combined method used in this paper is an advanced approach for two dipole sources localization with high accuracy and fast calculating.

In this paper, an intelligent controller is applied to electrically heated micro heat exchanger. First, the dynamics of the micro heat exchanger, which is a highly nonlinear plant, is identified using Locally Linear Model Tree (LOLIMOT) algorithm. Then, an intelligent controller is applied to the identified model. The performance of the proposed intelligent controller is compared with that of classic controllers like PID. Our results demonstrate excellent control action, disturbance handling and system parameter robustness for the intelligent controller.

Thework presented in this paper shows the capability of a connectionist model, based on a statistical technique called Exploratory Projection Pursuit (EPP), to identify anomalous situations related to the traffic which travels along a computer network. The main novelty of this research resides on the fact that the connectionist architecture used here has never been applied to the field of IDS (Intrusion Detection Systems) and network security. The IDS presented is used as a method to investigate the traffic which travels along the analysed network, detecting SNMP (Simple Network Management Protocol) anomalous traffic patterns. In this paper we have focused our attention on the study of two interesting and dangerous anomalous situations: a port sweep and a MIB (Management Information Base) information transfer. The presented IDS is a useful visualization tool for network administrators to study anomalous situations related to SNMP and decide if they are intrusions or not. To show the power of the method, we illustrate our research by using real intrusion detection scenario specific data sets.

Gaussian processes have been favourably compared to backpropagation neural networks as a tool for regression. We show that a recurrent neural network can implement exact Gaussian process inference using only linear neurons that integrate their inputs over time, inhibitory recurrent connections, and one-shot Hebbian learning. The network amounts to a dynamical system which relaxes to the correct solution. We prove conditions for convergence, show how the system can act as its own teacher in order to produce rapid predictions, and comment on the biological plausibility of such a network.

Gaussian mixture modelling is a powerful tool for data analysis. However, the selection of number of Gaussians in the mixture, i.e., the mixture model or scale selection, remains a difficult problem. In this paper, we propose a new kind of dynamic merge-or-split learning (DMOSL) algorithm on Gaussian mixture such that the number of Gaussians can be determined automatically with a dynamic merge-or-split operation among estimated Gaussians from the EM algorithm. It is demonstrated by the simulation experiments that the DMOSL algorithm can automatically determine the number of Gaussians in a sample data set, and also lead to a good estimation of the parameters in the original mixture. Moreover, the DMOSL algorithm is applied to the classification of Iris data.

Bayesian radial basis function neural network is presented to explore the weight structure in radial-basis function neural networks for discriminant analysis. The work is motivated by the empirical experiments where the weights often follow certain probability density functions in protein sequence analysis using the bio-basis function neural network, an extension to radial basis function neural networks. An expectation-maximization learning algorithm is proposed for the estimation of the weights of the proposed Bayesian radial-basis function neural network and the simulation results show that the proposed novel radial basis function neural network performed the best among various algorithms.

Racing algorithms have recently been proposed as a general-purpose method for performing model selection in machine learning algorithms. In this paper, we present an empirical study of the Hoeffding racing algorithm for selecting the parameter in a simple -nearest neighbor classifier. Fifteen widely-used classification datasets from UCI are used and experiments conducted across different confidence levels for racing. The results reveal a significant amount of sensitivity of the -nn classifier to its model parameter value. The Hoeffding racing algorithm also varies widely in its performance, in terms of the computational savings gained over an exhaustive evaluation. While in some cases the savings gained are quite small, the racing algorithm proved to be highly robust to the possibility of erroneously eliminating the optimal models. All results were strongly dependent on the datasets used.

Consider a network of unreliable links, each of which comes with a certain price and reliability. Given a fixed budget, which links should be bought in order to maximize the system’s reliability? We introduce a Cross-Entropy approach to this problem, which can deal effectively with the noise and constraints in this difficult combinatorial optimization problem. Numerical results demonstrate the effectiveness of the proposed technique.