Catálogo de publicaciones - libros

Feature selection for ensembles has shown to be an effective strategy for ensemble creation due to its ability of producing good subsets of features, which make the classifiers of the ensemble disagree on difficult cases. In this paper we present an ensemble feature selection approach based on a hierarchical multi-objective genetic algorithm. The algorithm operates in two levels. Firstly, it performs feature selection in order to generate a set of classifiers and then it chooses the best team of classifiers. In order to show its robustness, the method is evaluated in two different contexts: supervised and unsupervised feature selection. In the former, we have considered the problem of handwritten digit recognition while in the latter, we took into account the problem of handwritten month word recognition. Experiments and comparisons with classical methods, such as Bagging and Boosting, demonstrated that the proposed methodology brings compelling improvements when classifiers have to work with very low error rates.

With the invention of microarray technology, researchers are able to measure the expression levels of ten thousands of genes in parallel at various time points of a biological process. The investigation of gene regulatory networks has become one of the major topics in Systems Biology. In this paper we address the problem of finding gene regulatory networks from experimental DNA microarray data. We suggest to use a multi-objective evolutionary algorithm to identify the parameters of a non-linear system given by the observed data. Currently, only limited information on gene regulatory pathways is available in Systems Biology. Not only the actual parameters of the examined system are unknown, also the connectivity of the components is a priori not known. However, this number is crucial for the inference process. Therefore, we propose a method, which uses the connectivity as an optimization objective in addition to the data dissimilarity (relative standard error - RSE) between experimental and simulated data.

A large-scale, real-world application of Evolutionary Multi- Criterion Optimization (EMO) is reported in this paper. The Multidisciplinary Design Optimization among aerodynamics, structures and aeroelasticity for the wing of a transonic regional jet aircraft has been performed using high-.delity models. An Euler/Navier-Stokes (N-S) Computational Fluid Dynamics (CFD) solver is employed for the aerodynamic evaluation. The NASTRAN, a commercial software, is coupled with a CFD solver for the structural and aeroelastic evaluations. Adaptive Range Multi-Objective Genetic Algorithm is employed as an optimizer. The objective functions are minimizations of block fuel and maximum takeo. weight in addition to di.erence in the drag between transonic and subsonic .ight conditions. As a result, nine non-dominated solutions have been generated. They are used for tradeo. analysis among three objectives. One solution is found to have one percent improvement in the block fuel compared to the original geometry designed in the conventional manner. All the solutions evaluated during the evolution are analyzed by Self-Organizing Map to extract key features of the design space.

The optimization and design of two different types of photonic devices – a Fibre Bragg Grating and a Microstructured Polymer Optical Fibre is presented in light of multiple conflicting objectives in both problems. The fibre grating optimization uses a fixed length real valued representation, requiring the simultaneous optimization of four objectives along with variable bounds and a single objective constraint. This led to the human selection of a Pareto-optimal design which was manufactured. The microstructured fibre design process employs a new binary encoded variable length representation. An external embryogeny, or growth process is used to guarantee the creative generation of these complex designs which are automatically valid with respect to manufacturing constraints. Some initial results are presented for the case of two objectives which relate to the bandwidth and signal loss of a design.

The paper describes the application of multiobjective evolutionary algorithms in multicriteria optimization of operational production plans in a foundry, which produces iron castings and uses hand molding machines. A mathematical model that maximizes utilization of the bottleneck machines and minimizes backlogged production is presented. The model includes all the constraints resulting from the limited capacities of furnaces and machine lines, limited resources, customers requirements and the requirements of the manufacturing process itself. Test problems based on real production data were used for evaluation of the different evolutionary algorithm variants. Finally, the plans were calculated for a nine week rolling planning horizon and compared to real historical data.

We propose a new approach for multiobjective shape optimization based on the estimation of probability distributions. The algorithm improves search space exploration by capturing landscape information into the probability distribution of the population. Correlation among design variables is also used for the computation of probability distributions. The algorithm uses finite element method to evaluate objective functions and constraints. We provide several design problems and we show Pareto front examples. The design goals are: minimum weight and minimum nodal displacement, without holes or unconnected elements in the structure.

Due to the environmental concerns that arise from the emissions produced by fossil-fueled electric power plants, the classical economic dispatch, which operates electric power systems so as to minimize only the total fuel cost, can no longer be considered alone. Thus, by environmental dispatch, emissions can be reduced by dispatch of power generation to minimize emissions. The environmental/economic dispatch problem has been most commonly solved using a deterministic approach. However, power generated, system loads, fuel cost and emission coefficients are subjected to inaccuracies and uncertainties in real-world situations. In this paper, the problem is tackled using both deterministic and stochastic approaches of different complexities. The Nondominated Sorting Genetic Algorithm – II (NSGA-II), an elitist multi-objective evolutionary algorithm capable of finding multiple Pareto-optimal solutions with good diversity in one single run is used for solving the environmental/economic dispatch problem. Simulation results are presented for the standard IEEE 30-bus system.

The integrated management of financial risks represents one of the main challenges in contemporary banking business. Deviating from a rather silo-based approach to risk management banks put increasing efforts into aggregating risks across different risk types and also across different business units to obtain an overall risk picture and to manage risk and return on a consolidated level. Up to now no state-of-the-art approach to fulfill this task has emerged yet. Risk managers struggle with a number of important issues including unstable and weakly founded correlation assumptions, inconsistent risk metrics and differing time horizons for the different risk types. In this contribution we present a novel approach that overcomes parts of these unresolved issues. By defining a multi-objective optimization problem we avoid the main drawback of other approaches which try to aggregate different risk metrics that do not fit together. A MOEA is a natural choice in our multi-objective context since some common real-world objective functions in risk management are non-linear and non-convex. To illustrate the use of a MOEA, we apply the NSGA-II to a sample real-world instance of our multi-objective problem. The presented approach is flexible with respect to modifications and extensions concerning real-world risk measurement methodologies, correlation assumptions, different time horizons and additional risk types.

The vast majority of the developed planning methods for power distribution systems consider only one objective function to optimize. This function represents the economical costs of the systems. However, there are other planning aspects that should be considered but they can not be expressed in terms of costs; therefore, they need to be formulated as separate objective functions. This paper presents a new multi-objective planning method for power distribution systems. The method is based on the Strength Pareto Evolu-tionary Algorithm 2. The edge-set encoding technique and the constrain-domination concept were applied to handle the problem constraints. The method was tested on a real large-scale system with two objective functions: economical cost and energy non-supplied. From these results, it can be said that the proposed method is suitable to resolve the multi-objective problem of large-scale power distribution system expansion planning.

This paper deals with a two-objective rescheduling problem in a job shop for alteration of due date. One objective of this problem is to minimize the total tardiness, and the other is to minimize the difference of schedule. A genetic algorithm is proposed, and a new selection operation is particularly introduced to obtain the Pareto optimal solutions in the problem. At every generation in the proposed method, two solutions are picked up as the parents. While one of them is picked up from the population, the other is picked up from the archive solution set. Then, two solutions are selected from these parents and four children generated by means of the crossover and the mutation operation. The candidates selected are not only solutions close to the Pareto-optimal front but also solutions with a smaller value of the total tardiness, because the initial solutions are around the solution in which the total tardiness is zero. For this purpose, the solution space is ranked on the basis of the archive solutions. It is confirmed from the computational result that the proposed method outperforms other methods.