OPERATIONAL RESEARCH

Author: SISTIAGA LOPETEGI JUAN JOSE.
Year: 2004.
University: PAÍS VASCO.
Place of defense: FACULTAD DE PSICOLOGIA.
Place of preparation: FACULTAD DE PSICOLOGIA.

Summary: This is a research paper with a design "ex post facto", conducted with a methodology cross descriptive correlational and where we wanted to know the time to practice sports and current retrospective, from 9 to 16 years. We have also analyzed the level of perceived self-efficacy general and specifically to play football, and the level of satisfaction with the practice of football as psychological variables of high importance in the training of young football players. To this end we used a sample of 202 subjects between 12 and 18 years practicing football Land in the province of Guipuzcoa.

Author: MOLINA VILA MARÍA DOLORES.
Year: 2004.
University: ALICANTE.
Place of defense: FACULTAD DE CIENCIAS.
Place of preparation: FACULTAD DE CIENCIAS.

Summary: Programming Semi-Infinita Linear (PSIL) studied the problem of optimizing a linear function in Rn subject to a system of linear inequalities each of them associated with an index t T a set that can be infinite. The objective of this report is classified as excessive or necesaria.la information contained in the data from the linear programming problems (where T is finite and that abbreviated mind is denoted PL) and programming semi-infinita li ... (PSIL) which are called interchangeably linear optimization problems. It develops a theoretical framework that can be characterized and competent various phenomena of excess information appeared in the literature or defined here for the first time. It classifies a inecuación or correspondie variable dual (an index, in any case) as superfua or not with respect to six objects of interest means all practicable, the optimal set and the value of Optimal problems yearlings and why duales.La which authors have paid attention to information overload in the optimization problems affecting both their theoretical properties and the computational efficiency of the methods used to resolve numérica.EI Chapter O introductory incorporates notation, concepts and the results to be used in the future. This chapter also contains some preliminary results original to be used later. In Chapter 1 discusses rates superfluous to the problems not parametritos. Section 1.1 reviewed the literature on and complete redundancy (ie, on the conditions for the index to be redundant with respect to all practicable p1; imal), Section 1.2 incorporates the study of other 5 classes indexes superfluous, while Section 1.3 provides that the new results on the classical concept of saturation. In Chapter 2 examines the case for metric when the parameter is the vector object and runs throughout Rn. We study the six kinds of indexes superfluous, as well as two classical concepts such as surprise and inesencialidad. In Chapter 3 also considered and if parametrito, but the focus is on the restrictions evenly saturated. Somehow, it is almost never superfluida The restrictions Chapter 4, and finally brings together a collection of examples intended to separate the large number of interrelated concepts that appear in the previous chapter and to show the necessity of the assumptions in the results.

Author: MARTÍN BARRAGÁN BELÉN.
Year: 2005.
University: SEVILLA.
Place of defense: FACULTAD DE MATEMÁTICAS.
Place of preparation: FACULTAD DE MATEMÁTICAS.

Summary: The recent need to analyze large amounts of data of all kinds has led to the emergence, on the border of several deferent areas of the Data Mining. One of the problems facing is the classification, known as Classical Statistical Analysis Discriminante. In this thesis we propose the use of tools forefront of the Mathematical Programming classifiers to get more efficient, based on the Support Vector Machines. The problem of classification, the main objective is to obtain classified correctly classifiers that a high percentage of objects. However, it also has a strong interest that are also cheap, or easily interpretable, or useful to detect relevant variables, or take into account the importance of incorrectly classifying an object depends on the class to which it belongs. Over the different chapters of this thesis, we use different tools of the Mathematical Programming, as Generation of Columns, Programming Biobjetivo, or Mixed Integer Programming, to take account of these desirable properties in classifiers obtained. Our results, both theoretical and computational show the usefulness of such tools.

Author: PIZARRO ROMERO CELESTE.
Year: 2005.
University: REY JUAN CARLOS.
Place of defense: DEPARTAMENTO DE ESTADÍSTICA E INVESTIGACIÓN OPERATIVA.
Place of preparation: ESCET. UNIVERSIDAD REY JUAN CARLOS.

Summary: This thesis deals with the entire field of Stochastic Programming and raised new models and methods of resolution, both general and specific. Specifically, we study models that make use of binary variables to represent complex conditions. The resulting models are extremely difficult to solve because it combines the difficulties of the combinatorial model with the difficulties inherent in Stochastic Programming. Therefore, another objective of this work is to develop computationally efficient methods to get good quality solutions for these models.

Author: PAGES BERNAUS ADELA.
Year: 2006.
University: POLITÉCNICA DE CATALUÑA.
Place of defense: FACULTAT DE MATEMÀTIQUES I ESTADÍSTICA.
Place of preparation: ETSEIB, Edifici H PLANTA 6, DESPATX: 6.63 SD.

Summary: This thesis presents an approach to the long-term planning of power generation for a company (SGC) participating in a liberalized market organized as a pool. The goal of this thesis is two-fold: to model the problem and to develop and implement appropriate and efficient techniques for solving it. The optimization of the long-term generation planning is important for budgeting and planning fuel acquisitions, and to give a frame where to fit short-term generation planning. Our proposal for planning long-term generation is to optimize the expected generation of each unit (or the merger of several units of the same type) in the power pool over each interval into which the long-term horizon is split. The basic model for long-term generation planning (LTGP) maximizes the profit for all the units participating in the pool. The most important constraint is matching demand, since the market always clears. The Bloom and Gallant formulation is used, which models the load with a load-duration curve for each interval and requires an exponential number of linear inequality constraints, called herein LMCs. Other (linear) constraints included in the model are: minimum generation time, limits on the availability of fuel, maximum CO2 emission limits or the market share of the SGC. This thesis also proposes the way in which coordination between the LTGP model developed and a short-term plan should be considered and provides a model for short-term electrical power planning adapted to the LTGP proposed and which includes the long-term results. Another decision that needs to be taken from a long-term point of view is the joint scheduling of thermal unit maintenances with the generation planning of a particular SGC. The results of a prototype of a Branch and Bound implemented in AMPL are included in this thesis. Long-term planning needs to be considered before short-term planning and whenever the real situation deviates from the forecasted parameters, so the techniques implemented must be efficient so as to provide reliable solutions in a short time. Two methods for handling the LMCs are proposed and compared:  A decomposition technique exploits the fact that the LMCs plus the non-negativity bounds define a convex polyhedron for each interval whose vertices are easy to find. Thus, the problem is transformed and the variables become the coefficients of a convex combination of the vertices. The transformed problem is quadratic with linear constraints, making it suitable to be solved with the Murtagh & Saunders algorithm, which gives an optimal solution. A column-generation approach is used because the number of vertices of the polyhedron is comparable to the number of LMCs. The advantage of this method is that it does not require previous computation of all of the vertices, but rather computes them as the algorithm iterates.  The application of direct methods is computationally difficult because of the exponential number of inequality LMCs. However, only a reduced subset of LMCs will be active at the optimizer. A heuristic, named GP heuristic, has been devised which is able to find a reduced set of LMCs including those that are active at the optimizer. It solves a sequence of quadratic problems in which the set of LMCs considered is enlarged at each iteration. The quadratic problems are solved with an interior point method, and warm starts are employed to accelerate the solution of the successively enlarged quadratic problems. This procedure is more efficient than the column generation one. The modeling and tests of this thesis are based on the pure pool system and market data from the Spanish system up to July 2006.

Author: CANO BELMAN JAIME.
Year: 2006.
University: POLITÉCNICA DE CATALUÑA.
Place of defense: AULA 28.8 DE L'ETSEIB.
Place of preparation: ETSEIB, Edifici H PLANTA 6, DESPATX: 6.63 SD.

Summary: The object of this thesis is the Mixed-model Assembly-line Sequencing problem. Assembly lines have become a core component in modern repetitive manufacturing systems, especially in Just in Time production systems. Sequencing problems are important for the efficient use of mixed-model assembly lines. The sequencing for such lines entails the determination of the order in which the products are to be assembled in the line, in relation to specific criteria. In this work, load leveling is considered to be the sequencing criteria. Various objectives have been considered by researchers to measure leveling of the load. The objective used depends on the company management policies and on the productive system constraints. The main sequencing objective is to minimize the work overload. Work overload can be understood as the work which would not be completed if the assembler were to work at a normal pace and stay within the boundaries of the station. The problem is studied assuming paced line, closed stations, deterministic processing times, fixed launching, and operator schedule has an early start. Due to the computational effort required for optimally solving the problem, three different heuristic methodologies are explored in this thesis: constructive procedures (CP), local search and the combination of priority rules and neighborhood search. Concerning the CP, three algorithms are taken from literature. These procedures consider one station and two kinds of products . Two procedures are proposed for this problem. One of them is greedy (Ud1). The other (Ud2) is based in the fact that prior values of compulsory work overload and idle time can be obtained and can be used to guide previously existing procedures. Five procedures are compared in two computational experiments. In experiment one, procedures are compared on a testbed of 750 instances with single station. Multiple stations are considered in experiment two with 2250 instances . With regards to the single station problem, the best results are obtained with Ud2. For the multi station case, one of the literature procedures (YR) and Ud1 produce the best results. Since Ud1 and YR give the best results, the corresponding extensions are proposed which consider multiple products . Three versions of Ud1 are given and one of YR. The Ud1 extensions include concepts such as multi-cycles, system regeneration and product load smoothing. A experiment is used to test the efficiency of the procedures with instances obtained from the literature. The best results are obtained with one of the Ud1extensions, which tries to smooth out the load and idleness along the sequence. In the second part of this thesis, local search procedures are applied on the results obtained by the CP. Two kinds of neighborhoods are used: element exchange and segment insertion. In the experiment, a variety of exchange alternatives and various segment sizes are tested. Since neighborhoods are much larger when the considered moves are element exchange, search procedures based on segment insertion were faster and also obtain better solutions. The combination of two types of heuristics is tested in the last part of this thesis: the priority rules and the generation of new solutions in the search space. The hyperheuristic proposed (HH1) is based on the Scatter Search metaheuristic (SS). In greedy procedures, priority rules are used to establish an ordered list of candidate products to be sequenced at each decision stage. In this thesis, 20 priority rules are used. These rules consider processing times, pending demand, loads, displacements of workers in the stations, the bottleneck station, work overload, idle time, and smoothness in work overload and production. HH1 differs from SS in various aspects. As in SS, the HH1 also uses a reference set, but instead of solutions, it includes chains containing priority rules. By combining the chains of priority rules in the reference set, new ones are produced.

Author: HERNANDEZ MALDONADO MARIA LUISA.
Year: 2006.
University: POLITÉCNICA DE CATALUÑA.
Place of defense: SALA D'ACTES FME.
Place of preparation: ETSEIB, Edifici H PLANTA 6, DESPATX: 6.63 SD.

Summary: The study of multiple tables that tolerate groups of measured quantitative variables on a same set of individuals has provoked numerous methodologies in the field of the factorial analysis, sometimes extended for multiple tables of qualitative or mixed type. Less attention has been paid to the Multiple Contingency Tables (MCT) and is this the subject of the thesis. The main objectives of the thesis are: 1) to study the factorial methods for the treatment of the MCT; 2) to make a systematic comparison of these methods from as much theoretical point of view as applied; 3) to value the potential of application in the textual analysis. In order to reach these objectives it is necessary: 4) to elaborate a methodology for the systematic comparison of the methods. The study of the MCT presents two important problems: o the necessity to balance the influence of the different sub-tables in the global factorial analysis; o the difference between the margin-row from a sub-table to another one. First part Chapter 1 describes the General Factorial Analysis (GA), frame of reference for the factorial methods, and the Correspondence Analysis (CA), method of reference for the factorial analysis of contingency tables. Chapter 2 examines in a theoretical way the factorial methods for the study of the MCT: ⢠CA applied to the juxtaposed table, ⢠Intern CA (ICA), ⢠Multiple Factor Analysis for Contingency Tables (MFACT) ⢠Simultaneous Analysis for contingency tables (AS). The desirable properties are defined and it is shown, for each method, if they are fulfilled or not. This part is explained by means of small examples that emphasize the difficulties of the comparison. Chapter 3 shows the results obtained in the application of the different methods from a real example. The contribution of each method with the support of this example is compared systematically. Second part. Application to the textual analysis. Chapter 4 examines two selected legal corpora, for the treatment of multiple contingency Tables that contain data of textual type. The objectives of that study are also described. This second part shows two chronological corpora which include the sentences pronounced between 1979 and 1996 by the Supreme Court concerning two families of crimes. Chapter 5 shows the results obtained in the separated direct analyses, of said corpus. One of the objectives of these analyses is to extract the most contributing words, bases of the study of the later analyses. The stability of the selected words is verified. In Chapter 6, the sentences are regrouped per year, separately for each corpus. Two chronological corpora defined on the above-mentioned years are obtained. The separated analyses of correspondences of the lexical tables added âyears by wordsâ, allow to study the evolution of the vocabulary in each one of corpus (chronological study). The stability of the years is verified. Finally, in Chapter 7, the multiple lexical table that juxtaposes both corpora is studied, taking the year-row as the common dimension. The MFACT is applied to the MCT thus obtained, in order to carry out a comparative study of the chronology, looking for the similarities and differences between the vocabulary used in the corpora. The conclusion provides an assessment of the contribution of the thesis.

Author: FAJARDO GOMEZ MARIA DOLORES.
Year: 2006.
University: ALICANTE.
Place of defense: FACULTAD DE CIENCIAS.
Place of preparation: FACULTAD DE CIENCAS.

Summary: The Chapter 1 is devoted to the study of cualifación restrictions Locally Farkas-Mikowski (LFM) programming convex semi-infinita. It examines its relationship with the semicontinuidad higher in the sense Berge of the so-called multi-function active and supactiva. Certain condicones involving compliance with the LFM Properties guarantee regulate behavior of the supreme role of the functions involved in the system of restrictions on the problem and give validity to a formula for that kind Valadier supreme function. It also made a qualification restircciones type Slater which in turn implies qualification LFM. Chapter 2 deals with the study of the geometric properties of all solutions of a system convex semi-infinito, particularizando in systems LFM. The results obtained are compared with those already known in the linear case. As a first geometric problem solved caracaterización (partial) in terms of inclusion interior and the border (absolute and relative) set of solutions. Among the major differences with the geometric properties of linear systems consisting LFM highlights caracterizciones interior absolute and relative package of solutions, and proposes necessary and sufficient conditions that facilitate such characterizations. Chapter 3 introduces a new framework in which to study the properties of a function convex finito-valoada in terms of a system of linear inequalities which set of solutions is the epgirafo the role. This system is called representation of the function. Featured are three types of representations: LFM, Farkas-Minkowski (FM) and locally polyhedral (LOP). The existence of the latter kind of representation is derived the concept of function cuasipoliédrico. It generalizes the function poliédrica, being this kind of functions closed regarding conjugation, and major operations known, it is likewise about the roles cuasipoliédricas. You also get characterizations of subdiferencial and epsilon-subdiferencial a function cuasipoliédrica at one point. As implementation of the convexity cuasipoliédrica, is built around a convex space from a convex but not continuous.

Author: Barrios Sarmiento Agustín.
Year: 2006.
University: VALENCIA.
Place of defense: Facultad de Ciencias Matemáticas.
Place of preparation: Facultad de Ciencias Matemáticas.

Summary: This thesis addresses two major problems in the sequencing of projects with limited resources and timing relationships both as a minimum rate of maximum rate, they are, cases and unique way so multiple. In addition to his academic wealth, these problems arouse much interest given that allow you to create models that represent a wide range of real-life situations, including: manufacturing batch maintenance of complex systems, building infrastructure, and so on. For its complexity, we need efficient algorithms in terms of computational power to solve them. It introduces new concepts and techniques that can be used by other algorithms. It proposes an evolutionary algorithm, EVA, to resolve the problem so unique. Two algorithms are also proposed for a multiple case, the former is a double genetic algorithm, based on the method of integration, and the second BMS an algorithm based on the method of decomposition. The computational algorithms provide excellent results, which are compared with the results of the best algorithms published until the time of writing this thesis.