Department of Statistics and Operations Research

Seminar Schedule: Year 2007


PRICING IN COMBINATORIAL AUCTIONS

CONVIDAT: Andreas Drexl. Faculty of Economics and Business Administration, Christian-Albrechts-University of Kiel, Germany.

IDIOMA: Anglès

LLOC: Aula C5016, Campus Nord, UPC (veure mapa)

DATA: Dimecres, 14 de febrer de 2007, 11:30

RESUM: In combinatorial auctions the pricing problem is of main concern since it is the means by which the auctioneer signals the result of the auction to the participants. In order for the auction to be regarded as fair the price signals should be such that a participant that has won a subset of items knows why his bid was a winning bid and that agents that have not acquired any item easily can detect why they lost. The problem is that the winner determination problem is a hard integer programming problem and hence in general there do not exist linear prices supporting the optimal allocation. From IP duality theory we know that there exist non-linear anonymous price functions that support the optimal allocation. In this paper we will provide a means to obtain a simple form of a non-linear anonymous price system that supports the optimal allocation.

TRANSPARENCIES DE LA PRESENTACIO: clica aquí

Inici
MUESTREO Y ESTIMACIÓN EN UNA ENCUESTA DE EVALUACIÓN DE UN PROGRAMA DE DISTRIBUCIÓN DE LA RENTA EN BRASIL

CONVIDAT: Pedro Silva. Southampton Statistical Sciences Research Institute, University of Southampton, UK.

IDIOMA: Castellà.

LLOC: Aula C5016, Campus Nord, UPC (veure mapa)

DATA: Divendres, 9 de març de 2007, 12:30

RESUM: El programa "Bolsa familiar" es un programa de transferencia de renta del Gobierno Federal Brasileño a las familias en situación de indigencia o pobreza. Se presenta una descripción de los métodos de muestreo y ponderación utilizados en la encuesta domiciliaria, de ámbito nacional, realizada en noviembre 2005 como una de las acciones destinadas a evaluar el impacto de dicho programa sobre las familias. Se comentan algunos de los desafíos enfrentados y su relación con el estado del arte en los métodos de muestreo y estimación.

(Trabajo conjunto con Maurício Teixeira Leite de Vasconcellos (SCIENCE) y Nuno Duarte da Costa Bittencourt (SCIENCE)).

TRANSPARENCIES DE LA PRESENTACIO: clica aquí

Inici
TRANSPORTATION PROBLEMS AND INTEGER PROGRAMMING

CONVIDAT: Martin Grötschel. Konrad-Zuse-Zentrum fuer Informationstechnik Berlin (ZIB), Technische Universität Berlin, Institut für Mathematik, Germany.

IDIOMA: Anglès

LLOC: Aula C5016, Campus Nord, UPC (veure mapa)

DATA: Divendres, 16 de març de 2007, 11:30

RESUM: Transportation problems occuring in practice (such as: public mass transport by bus, train, or plane , dial-a-ride, truck routing, scheduling of service vehicles, logistics, ...) are much more complex tasks than what is usually called a "transportation problem" in the mathematical programming terminology. In this talk I will present an overview of some of the problems of this type that have been attacked using integer programming techniques by the transportation research group at the Konrad Zuse Zentrum and the DFG Research Center Matheon in Berlin. I will present solutions of very large scale instances from practice and indicate the savings that can result from the use of advanced mathematical solution technology. In my computational report I will particularly focus on vehicle scheduling and driver assignment in public transport.

Inici
LOT-SIZES FOR A FAMILY OF ITEMS PRODUCED ON A SINGLE MACHINE WITH VARIABLE PRODUCTION RATES

CONVIDAT: Oyvind Halskau. Molde University College, Norway.

IDIOMA: Anglès

LLOC: Aula C5016, Campus Nord, UPC (veure mapa)

DATA: Dilluns, 26 de març de 2007, 11:30

RESUM: Over the last years production technology has become more flexible. A single machine can produce different items both in sequence and in parallel, internal and external set up times and costs have been reduced, and computers are used in order to govern the production on such machines. In this paper the classical lot size problem of producing a family of items in a given sequence on a single machine with given production rates is revisited. It is assumed that the set up costs are independent of the sequencing of the production of the items, but in stead of assuming fixed production rates for the different items, these rates can be varied within certain limits. Procedures for finding sets of optimal production rates in different situations are offered. It turns out that maximal production rates for all the items are not an optimal solution in every case. Further, the sequencing and optimal production rates will depend on parameters like demands during time horizon, unit values and inventory holding costs for the items.

TRANSPARENCIES DE LA PRESENTACIO: clica aquí

Inici
SINGLE VEHICLE PICKUP AND DELIVERY PROBLEM WITH CAPACITATED CUSTOMERS

CONVIDAT: Irina Gribkovskaia. Molde University College, Norway.

IDIOMA: Anglès

LLOC: Aula C5016, Campus Nord, UPC (veure mapa)

DATA: Dimecres, 28 de març de 2007, 11:30

RESUM: We introduce a pickup and delivery problem encountered in the servicing of offshore oil and gas platforms in the Norwegian Sea. A single vessel must perform pickups and deliveries at several offshore platforms. All delivery demands originate at a supply base and all pickup demands are also destined to the base. The vessel capacity may never be exceeded along its route. In addition, the amount of space available for loading and unloading operations is limited at each platform. The problem, called the Single Vehicle Pickup and Delivery Problem with Capacitated Customers (SVPDPCC) consists of designing a least cost vehicle (vessel) route starting and ending at the depot (base), visiting each customer (platform), and such that there is always sufficient capacity in the vehicle and at the customer location to perform the pickup and delivery operations. A mixed integer linear programming model yielding general solutions allowing for up to two visits at customers is proposed, and the impact of the new stipulations on the solutions is analyzed. We have developed several construction heuristics as well as a tabu search algorithm. Our best heuristic consists of applying tabu search starting with an initial solution generated with construction heuristic which combines nearest neighbour, cheapest insertion and backward merging procedures. The solutions produced by our algorithm may be Hamiltonian or not, depending partly on vertex demands and relative locations.

TRANSPARENCIES DE LA PRESENTACIO: clica aquí

ARTICLE DE LA PRESENTACIO: clica aquí

Inici
SOLVING THE UNCAPACITATED FACILITY LOCATION PROBLEM WITH SEMI-LAGRANGIAN RELAXATION

CONVIDAT: Cèsar Beltran. Departamento de Estadística e Investigación Operativa, Universidad Rey Juan Carlos (Madrid).

IDIOMA: Català

LLOC: Aula C4002, Campus Nord, UPC (veure mapa)

DATA: Dimarts, 10 d'abril de 2007, 11:30

RESUM: The semi-Lagrangian Relaxation (SLR) method has been introduced in Beltran et al. (2006) to solve the p-median problem. In this paper we apply the method to the Uncapacitated Facility Location (UFL) problem. We perform computational experiments on two main collections of UFL problems with unknown optimal values. On one collection, we manage to solve to optimality 16 out of the 18 instances. On the second collection we solve 2 out of 18 instances and improve the Lagrangian lower bound. In this way, we prove that the Hybrid Multistart heuristic of Resende et al. (2006) provides near optimal solutions.

ARTICLE DE LA PRESENTACIO: clica aquí

Inici
INTRODUCCIÓ ALS PROCESSOS PUNTUALS I LES SEVES APLICACIONS

CONVIDAT: Carles Comas. Centre Tecnològic Forestal de Catalunya i Departament de Matemàtiques, Universitat Jaume I de Castelló.

IDIOMA: Català

LLOC: Aula C5016, Campus Nord, UPC (veure mapa)

DATA: Divendres, 13 d'abril de 2007, 12:30

RESUM: Un procés puntual és un mecanisme estocàstic que genera un conjunt de punts {x} contingut en una regió finita A. Exemples d'aquests processos apareixen en multitud de sistemes incloent les emissions de partícules des d'una font radioactiva, polsos elèctrics, seqüències naturals en el temps i la distribució d'arbres i d'animals en el seu medi. La teoria dels processos puntuals permet l'estudi de l'ocurrència d'aquests punts (partícules, individus) tant en l'espai com en el temps. En aquest seminari es vol introduir els aspectes bàsics de la teoria dels processos puntuals així com posar de manifest les aplicacions més importants d'aquesta branca de l'estadística-matemàtica. S'introduiran els conceptes bàsics que caracteritzen un procés puntual tal com la funció de distribució i les característiques de primer i segon ordre, així com la presentació dels models puntuals més importants i les noves aproximacions teòriques per l'estudi de sistemes continus en l'espai i el temps. Paral·lelament es farà èmfasis en les possibles aplicacions pràctiques de totes aquestes eines matemàtiques. El seminari acaba amb unes conclusions i un llistat bibliogràfic. 

TRANSPARÈNCIES DE LA PRESENTACIO: clica aquí

Inici
TOWARDS INTEGRATED MICRO-SCOPIC TRAFFIC AND INSTANTANEOUS EMISSION MODELS

CONVIDAT: James E. Tate. Institute for Transport Studies, The University of Leeds

IDIOMA: English

LLOC: Aula C5016, Campus Nord, UPC (veure mapa)

DATA: Dimecres, 18 d'abril de 2007, 11:30

RESUM: Researchers and policy makers at local levels are demanding improved modelling tools to predict vehicle fuel consumption and emissions. By integrating traffic microsimulation and instantaneous emission models, tools can be developed that are capable of assessing the environmental impact of road transport policies and management strategies. Modern engine technology and control devices have changed the emission characteristics of vehicles, with tail-pipe emissions of CO, NOX, VOCs and Particles now largely associated with significant acceleration events. This, alongside developments in instrumentation technology, has led to a change in the emphasis of research and legislative emission testing procedures from steady state (constant engine speed and load) integrated measurements, through to transient or second-by-second testing using rolling road, chassis dynamometer or in-vehicle instrumentation. Using illustrative transient emission measurements the key factors affecting vehicular emissions are firstly reviewed, including the influence of: aggressive driving, vehicle fleet composition, cold starts and variations in driver behaviour. Passenger car fuel consumption and emissions are usually assessed as a function of average speed along a road section. As tail-pipe emissions from modern vehicles are now largely associated with acceleration events, the average speed approach lacks the resolution to be able to predict changes in pollutant emissions as traffic management policies influence driving patterns. Traffic microsimulation models however are able to predict the progression and behaviour of individual vehicles through a well-defined road network. When combined with an instantaneous emission model, the magnitude and location of acceleration and emission events can be predicted. The instantaneous 'California Modal Emission Model' (CMEM) emission model has been integrated with two traffic microsimulation models, namely DRACULA and AIMSUN NG. The integrated modelling approach has been applied over a small well calibrated and validated network in the City of York. The integrated modelling demonstrated that vehicular emissions of CO and NOX, largely associated with positive accelerations are concentrated around the intersection in free-flowing traffic conditions. When traffic queues build up away from the intersection emissions are elevated at mid-link locations, but are still significantly lower than at the intersection. Therefore commonly adopted air quality modelling procedures that assume vehicular emissions are evenly distributed along a road section are likely to under-predict emissions of traffic related pollutants in the vicinity of intersections by a factor approaching 5, whilst over-estimating emissions produced at mid-link locations. These findings suggest that integrated local-scale air quality modelling studies should consider all aspects at the microscopic level in order to unravel the complex inter-relationships between traffic, emissions, dispersion processes and resultant effects on air quality and exposure at the local-scale. Improved passenger car instantaneous fuel consumption and emission models in development, being constructed with comprehensive real-world measurements, will also be introduced.

Inici
STATISTICAL TOOLS IN KANSEI ENGINEERING

CONVIDAT: Simon Schutte. Linkopig University, Suecia.

IDIOMA: Anglès

LLOC: Aula C5016, Campus Nord, UPC (veure mapa)

DATA: Divendres, 20 d'abril de 2007, 12:30

RESUM: The methodology of Kansei Engineering has its origins in the 197ies. The Japanese Professor Mitsuo Nagamachi (1989) developed a method which he called "emotional Engineering". It was able to quantify the affective impact of products on the human mind and translate them into concrete product solutions. Later, when other research groups joined research on the topic it eventually changed name to "Kansei Engineering". From the start researchers used mathematical statistical tools for quantification to build models in order to predict emotional reactions of customers on products. The seminar will give a quick overview on Kansei Engineering structure itself. It focuses in particular on how the linkage between affective meaning and product properties is done and which methods are used. One of the earliest methods was Osgood's (1957) theory of the "Semantic Space"; explaining how humans describe products verbally. Soon Hiyashi's "Quantification Theory Type I" (1976) was used in order to link the semantic descriptions to product properties. The cutting edge of Kansei Engineering today is fuzzy logics, generic algorithm and rough set analysis.

Inici
LIMITED AND FULL INFORMATION ESTIMATION AND GOODNESS-OF-FIT TESTING IN MULTIDIMENSIONAL CONTINGENCY TABLES: A UNIFIED FRAMEWORK

CONVIDAT: Alberto Maydeu. Facultad de Psicologia, Universidad de Barcelona.

IDIOMA: Castellà

LLOC: Aula C5016, Campus Nord, UPC (veure mapa)

DATA: Divendres, 27 d'abril de 2007, 12:30

RESUM: High-dimensional contingency tables tend to be sparse, and standard goodness-of-fit statistics such as X^2 cannot be used without pooling categories. As an improvement on arbitrary pooling, for goodness of fit of large 2^n contingency tables, we propose classes of quadratic form statistics based on the residuals of margins or multivariate moments up to order r. These classes of test statistics are asymptotically chi-squared distributed under the null hypothesis. Further, the marginal residuals are useful for diagnosing lack of fit of parametric models. We show that when r is small (r = 2, 3), the proposed statistics have better small-sample properties and are asymptotically more powerful than X^2 for some useful multivariate binary models. Related to these test statistics is a class of limited-information estimators based on low-dimensional margins. We show that these estimators have high efficiency for one commonly used latent trait model for binary data.

ARTICLE EN EL QUE ES BASA LA PRESENTACIO: Maydeu & Joe (2005) JASA

Inici
EFFICIENT IMPLEMENTATION TECHNIQUES OF INTERIOR POINT METHODS FOR LARGE-SCALE OPTIMIZATION

CONVIDAT: Csaba Mészáros. Laboratory of Operation Research and Decision Systems, Computer and Automation Research Institute, Hungarian Academy of Sciences, Budapest, Hungary.

IDIOMA: English

LLOC: Aula C5016, Campus Nord, UPC (veure mapa)

DATA: Divendres, 11 de maig de 2007, 11:30.

RESUM: In 1984 Karmarkar presented a novel interior point method (IPM), which, as he claimed, was able to solve large-scale linear programs significantly faster than the simplex algorithm. This statement led to an explosion of interest in IPMs among researchers and practitioners. Indeed, the above resulted in an impressive progress in the theory and implementation technology of IPMs during the past 2 decades. It is now widely accepted that IPMs are numerically reliable optimization tools and on large-scale problems they can significantly outperform the algorithms based on the simplex approach. In the talk we give an overview about the state-of-the-art implementation techniques of IPMs. Since the practical success of any IPM implementation depends on the efficiency and the reliability of the linear algebra kernels in it, these issues will be in the focus of our survey. We will discuss algorithms related to the efficient exploitation of sparsity and methods related to the numerical computations. A special emphasis will be given on the implementation techniques that exploit recent hardware architectures such as hierarchical memory structures, vector and parallel processing capabilities. The most relevant points of IPM implementations will be demonstrated by computational results to illustrate typical behavior of the presented implementation techniques.

TRANSPARÈNCIES DE LA PRESENTACIO: clica aquí

Inici
DAY-AHEAD BIDDING AT NORDPOOL USING STOCHASTIC PROGRAMMING

CONVIDAT: Stein-Erik Fleten. Department of Industrial Economics and Technology Management Norwegian University of Science and Technology Trondheim, Norway.

IDIOMA: English

LLOC: Aula C5016, Campus Nord, UPC (veure mapa)

DATA: Divendres, 25 de maig de 2007, 12:00.

RESUM: From the point of view of a price-taking hydropower producer participating in the day-ahead power market, market prices are highly uncertain. This work provides a model for determining optimal bidding strategies taking this uncertainty into account. In particular, market price scenarios are generated and a stochastic mixed-integer linear programming model that involves both hydropower production and physical trading aspects is developed. The idea is to explore the effects of including uncertainty explicitly into optimization by comparing the stochastic approach to a deterministic approach. The model is illustrated with data from a Norwegian hydropower producer and the Nordic power market at Nord Pool.

Inici
A MULTIVARIATE BAYESIAN SPATIAL MODELING FRAMEWORK FOR HURRICANE SURFACE WIND FIELDS

CONVIDAT: Montserrat Fuentes. Statistics Department at North Carolina State, USA.

IDIOMA: Spanish.

LLOC: Aula 005 FME, UPC

DATA: Dimarts, 5 de juny de 2007, 13:00.

RESUM: Storm surge is the onshore rush of sea water caused by the high winds and to a lesser extent the low pressure associated with a hurricane. Storm surge can compound the effects of inland flooding caused by rainfall, leading to loss of property and loss of life for residents of coastal areas. Numerical ocean models are essential for creating nowcast estimates as well as forecasts for coastal areas that are likely to be impacted by the storm surge. These models are driven primarily by the surface wind forcings. Currently, gridded wind fields used to spin up and force ocean models are specified by deterministic formulas that provide an idealized form of the wind profile based on the central pressure and location of the storm center. While these equations incorporate important physical knowledge about the structure of hurricane surface wind fields they cannot always capture the asymmetric and dynamic nature of a hurricane. A new Bayesian multivariate spatial temporal statistical modeling framework is introduced to improve the estimation of the wind field inputs. A nonstationary and nonseparable spatial-temporal linear model of coregionalization (LMC) is developed and applied to explain the spatial-temporal variability in the wind vectors (u and v components), as well as the cross-dependency between these two components. This Bayesian framework allows for estimation of the parameters of the multivariate spatial model and the physically based wind model while accounting for potential additive and multiplicative bias in the observed wind data from buoys, ships, aircraft and satellite data. We find that this multivariate spatial model consistently improves parameter estimation and prediction for surface wind data for a case study of Hurricane Charley (2004) when compared to the original physical model. These methods are also shown to improve storm surge estimates when used as the forcing fields for a numerical three dimensional coastal ocean model.

Inici
TWO NONLINEAR TIME SERIES MODELS

CONVIDAT: David A. Dickey. Statistics Department at North Carolina State, USA.

IDIOMA: English

LLOC: Aula 005 FME, UPC

DATA: Dijous, 7 de juny de 2007, 13:00.

RESUM: We discuss two applications of the logistic function in time series. In the first, we look at a transfer function model for some stream flow data. Using a logistic function of lagged flow, we allow the lag relationship between the upstream and downstream stations to vary over time. In the second application, we look at properties of the autoregressive order one process with a hyperbolic tangent function of the lagged response replacing the usual fixed lag 1 coefficient. This function is just a logistic function linearly stretched to span the interval (-1,1), the range of stationary values. Because the estimate of the autoregressive order 1 coefficient has quite a different distribution, in the fixed coefficient case, for coefficients at or near 1 than it does for coefficients well inside the stationarity region, interest centers on the distributional behavior of the resulting estimators. A logistic function can be used if the interval (0,1) is of interest.

Inici
MODELIZACIÓN LINEAL JERÁRQUICA: UNA VISIÓN DE SUS ORÍGENES Y PERSPECTIVAS DE APLICACIÓN

CONVIDAT: Mario Miguel Ojeda. Universidad Veracruzana, México.

IDIOMA: English

LLOC: Seminario EIO, planta 6, ETSEIB (Edifici Eng. Industrial), Avda. Diagonal, 647.

DATA: Divendres, 8 de juny de 2007, 12:30.

RESUM: Se presentará una revisión de los orígenes y desarrollos de la modelización lineal jerárquica hasta nuestros días, enfatizando en las posibilidades de aplicación de esta metodología. Se ilustrará con algunos casos concretos de aplicación de los llamados modelos multinivel lineales. Se charlará sobre las generalizaciones y las áreas de oportunidad para los desarrollos metodológicos.

Inici
GRASP PARA REDES NEURONALES

CONVIDAT: Francisco Ángel-Bello Acosta. Instituto Tecnológico de Monterrey, México.

IDIOMA: Castellano

LLOC: Aula C5016, Campus Nord, UPC (veure mapa)

DATA: Dilluns, 9 de juliol de 2007, 12:00.

RESUM: En esta charla se aborda el problema de diseño y entrenamiento de redes neuronales artificiales con funciones de activación discretas para la clasificación de patrones en dos categorías. Cuando el conjunto de patrones no es linealmente separable, este problema consiste en determinar la cantidad de neuronas en la capa oculta que se requieren para clasificar correctamente los patrones, se sabe que este problema es NP-hard. En este caso, se propone un GRASP el cual explota la estructura particular del modelo para determinar las neuronas de la capa oculta así como los pesos correspondientes. Este procedimiento añade neuronas, una en cada paso, hasta que no quedan patrones por clasificar. Entonces es posible aplicar la condición de separabilidad lineal para obtener los correspondientes pesos de las neuronas. Como resultado se obtiene una red entrenada la cual clasifica todos los patrones en el conjunto de entrenamiento. El procedimiento se prueba con diez conjuntos benchmark y los resultados demuestran un buen desempeño en una cantidad de tiempo razonable.

Inici
VARIABLE SELECTION IN REGRESSION

CONVIDAT: Elías Moreno. Universidad de Granada.

IDIOMA: Castellano

LLOC: Aula C5016, Campus Nord, UPC (veure mapa)

DATA: Divendres, 23 de novembre de 2007, 12:30.

RESUM: (Joint work with George Casella, University of Florida; JASA 2006) A novel fully automatic Bayesian procedure for variable selection in normal regression model is proposed. The procedure is based on the model posterior probabilities and a computational strategy for model posterior evaluation. To compute model posterior probabilities, prior distributions for models and model parameters are needed. Uniform priors are assigned to models as usual, and intrinsic priors to model parameters. Intrinsic priors are default priors for model selection problems that are derived from the model structure, and do not contain any tuning parameter to be adjusted. So they can be seen as objective priors for variable selection.

Inici
LOCATION PROBLEMS IN SUPPLY CHAIN MANAGEMENT

CONVIDAT: Stefan Nickel. Chair of Operations Research and Logistics, Saarland University and Fraunhofer Institute for Industrial Mathematics.

IDIOMA: English

LLOC: Aula C5016, Campus Nord, UPC (veure mapa)

DATA: Dimecres, 28 de novembre de 2007, 12:00.

RESUM: Structuring global supply chain networks is a complex decision making process. The typical inputs to such a process consist of a set of customer zones to serve, a set of products to be manufactured, shipped and sold, demand projections for the different customer zones, information about future conditions and costs (e.g. transportation and production) and resources (e.g. capacities, available materials). Given the above inputs, companies have to decide, among other things, where to locate new service facilities (e.g. plants, warehouses), how to allocate procurement and production activities to the various manufacturing facilities of the network, and how to manage the distribution of products. We propose a mathematical modeling framework capturing many practical aspects of network design problems simultaneously that have not received adequate attention in the literature. The aspects considered include: dynamic planning horizon, generic supply chain network structure, external supply of materials, inventory opportunities for goods, distribution of commodities, facility configuration, availability of capital for investments, and storage limitations. Moreover, network configuration decisions concerning the gradual relocation of facilities over the planning horizon are considered. To cope with fluctuating demands, capacity expansion and reduction scenarios are also analyzed as well as modular capacity shifts. The relation of the proposed modeling framework with existing models is discussed. For problems of reasonable size we report on our computational experience with standard mathematical programming software. In particular, useful insights on the impact of various factors on network design decisions (like number of time periods) are provided. Moreover, solution approaches are presented using Lagrangian Relaxation and Benders Decomposition. Also a specially designed heuristic approach is proposed. This heuristic approach explores the solution of the linear relaxation of the problem. It successively rounds the fractional variables corresponding to the 0/1 decisions of changing the facilities' status (i.e., open new / close existing facilities), and it is also used to roughly estimate the total number of facility configuration changes over the planning horizon. The proposed heuristic performs very well on a large set of randomly generated problems. Some of the proposed models have been integrated into commercial software packages.

TRANSPARENCIES DE LA PRESENTACIO: clica aquí

Inici
PARTIALLY OBSERVED STOCHASTIC EPIDEMICS

CONVIDAT: Victor M. Panaretos. Institute of Mathematics, Ecole Polytechnique Fédérale de Lausanne, Suiza.

IDIOMA: English

LLOC: Sala de Juntes de la FME, UPC.

DATA: Dijous, 29 de novembre de 2007, 15:30.

RESUM: The modelling of the initial stages of a stochastic epidemic is often based on the assumption that the epidemic is fully observable. In practice, however, full observation is rarely possible, and this is something to be taken into account when making inferences concerning the explosion or extinction of such an epidemic. With such a situation in mind, we consider the probabilistic evolution of a discrete time branching process given its partial observation, that is, given a random "trace" of the process, which not only trails the process, but also directly affects it (in terms of interventions). The model proposed remains tractable, yielding reasonable closed form expressions, while preserving the marginal branching property of the unobservable true process.

Inici
RELIABILITY VS. TRAVEL DISTANCE TRADE-OFFS IN FACILITY LOCATION PROBLEMS

CONVIDAT: Mozart Menezes. Operations Management and Logistics Department HEC School of Management, Paris

IDIOMA: English

LLOC: Aula C6003, Campus Nord, UPC (veure mapa)

DATA: Dimecres, 3 de desembre de 2007, 12:00.

RESUM: In this discussion, we analyze a facility location model in a network where facilities may be subject to disruptions, causing customers to seek service from the operating facilities. We generalize the classical p-median problem on a network to explicitly include the failure probabilities, and analyze structural and algorithmic aspects of the resulting model. The optimal location patterns are seen to be strongly dependent on the probability of facility failure, with facilities becoming more centralized, or even co-located, as the failure probability grows. Several exact and heuristic solution approaches are developed. Results of numerical experiments are reported. In order to better view the effects of concentration and co-location, we shift methodologies and investigate a continuous location problem in a unit line. Effects found in the network can be better demonsstrated in this setting. Optimal solutions found in this simple case help us to quantify the importance of several inputs to the model. In the end, we will see a small case study where the city of Toronto and its hospitals are used to understand more the impact of disruption in the supply chain design.

 Inici
OPTIMAL WORKERS ALLOCATION FOR A CROSSDOCKING - JUST IN TIME SCHEDULING PROBLEM

CONVIDAT: José Luis González Velarde. Chair in Supply Chain, Inst. Tecnológico de Monterrey, México.

IDIOMA: Castellà.

LLOC: Aula C5016, Campus Nord, UPC (veure mapa)

DATA: Divendres, 14 de desembre de 2007, 12:00.

RESUM: In this talk, a warehouse is operated as a crossdock with a goal of minimizing total operating costs. These costs are due to two factors: the number of teams of workers hired to do the job, and the transit storage time for cargo. Each team of workers has a fixed cost per working day, and the cargo can incur earliness and tardiness delivery costs. The transit storage time for cargo is minimized according to Just in Time (JIT) scheduling. An integrated model to obtain both the optimal number of teams of workers and the schedule for the problem is formulated. Since the problem is known to be NP-hard, a solution approach (RGLSTS) based on a combination of two metaheuristics, Reactive GRASP embedded in a Local Search algorithm and Tabu Search is devloped. The results obtained from a MIP formulation solved with CPLEX for 14 problem instances and the results obtained from the RGLSTS metaheuristic algorithm for the same problem instances are discussed.
Inici