Elevator pitches — vrijdag 15 april

Voorzitter: Julia Mikhal (Universiteit Twente) Waaier 2, vr 12:45-13:15

Masterstudenten en promovendi presenteren hun onderzoek via posters en lichten deze toe in een presentatie van twee minuten.

Decentralized energy generation: a column generation approach

Maurice G.C. Bosman (Universiteit Twente) Waaier 2, vr 12:15-12:18

The ongoing shift towards a decentralized energy supply chain asks for new tools and methods to predict, plan and control the generation, storage and consumption of energy. The existing energy world needs to incorporate a large variation of distributed, smaller-scale generation technologies, which increase both the size and the complexity of the energy management problem, which in its simplest form is to realtime match supply and demand with limited storage possibilities. A column generation approach is developed to plan the production commitment of a fleet of decentralized energy generators, taking into account bounds on the total energy profile. This approach separates the local constraints on the production profiles from the global constraints on the total production to simplify the planning problem. The approach is able to compute solutions of relatively good quality in reasonable computation time.

Parameterization of atmospheric convection with conditional Markov chains

Jesse Dorrestijn (CWI) Waaier 2, vr 12:18-12:21

The effect of clouds and convection on the state of the atmosphere is a major source of uncertainty in weather and climate models. Explicit modeling of convection requires model resolutions on the order of 50 meters, whereas global climate models have horizontal resolutions starting at ~100 km. As a consequence, the effects of convection in individual vertical model columns must be represented (“parameterized”) in a simplified yet adequate way. We use a new method to parameterize convection stochastically. A Markov chain is constructed by using realistic data obtained from large-eddy simulation (LES) of atmospheric convection. With a cluster method representative vertical heat and moisture flux profiles are found and the corresponding transition probability matrix is estimated by counting the number of transitions between profiles. By conditioning the probabilities on the atmospheric state, a conditional Markov chain (CMC) is obtained that can produce realistic turbulent fluxes. These fluxes are crucial for convection parameterization in weather and climate models.

Online Outlier Detection in Testing of Integrated Circuits

Harm Bossers (Universiteit Twente) Waaier 2, vr 12:21-12:24

Testing of Integrated Circuits (IC’s) consists of measurements of all kind of values such as speed or leakage currents (IDDQ). The measurements need to be within certain specification limits, but these limits are usually quite wide to accommodate for variations caused by the production process or tester condition. Therefore, manufacturers want to apply outlier detection, since the outliers are potentially unreliable as is shown by empirical evidence. An outlier is defined as a measurement which differs significantly from an expected pattern of behavior, but is still within specification limits. However, most outlier detection methods are used in an offline setting and hence are not applicable to the Final Test stage, where immediate pass/fail decisions are required. So we need an online outlier detection method, since measurement distributions can shift due to all kind of variations. We developed a univariate online outlier detection method that is applicable to Final Test. Test limits are constructed based on previous measurements and updated with a rolling horizon. Robust statistics are used to ensure a stable start to the method. We analyzed our method using real-world data. We identified some cases which can result in performance degradation, but most experiments showed that our method is robust to outliers and able to detect them in an online setting. Furthermore, we show some work in progress about a bivariate online outlier detection method. This method is also a rolling horizon method, but it uses kernel density estimation to distinguish between “normal” (dense) regions and outlier regions.

Asymptotic behaviour of second order linear autonomous neutral delay differential equations

Guiling Chen (Universiteit Leiden) Waaier 2, vr 12:24-12:27

We present the asymptotic behaviour of the following second order linear autonomous neutral delay differential equation by ODE approach and spectral approach,

where a, b, c are real numbers, ( and ) are positive real numbers. The main idea of ODE approach is that of transforming the second order delay differential equation into first order delay differential equation, by using of a real root of the corresponding characteristic equation, while spectral approach is emphasis on the explicit computation of the large time behaviour by using spectral projections. Furthermore, the conditions given by the ODE approach are further studied and some examples are shown to illustrate the main results of this paper. The main results are based on the work of Driver’s asymptotic behaviour for first order delay differential equations and Frasson and Verduyn Lunel’s spectral theory for functional differential equations.

Balancing walk-in and appointments in health care

Nikky Kortbeek (Universiteit Twente) Waaier 2, vr 12:28-12:31

Outpatient and diagnostic testing clinics have long provided patients with appointments, to match capacity with demand. However, the main disadvantage of a pure appointment policy is that substantial access delays can be created. This study explores the viability of a walk-in based policy: a mixed strategy of walk-in and appointments. We present a stochastic method that finds the mixed strategy that achieves an optimal balance between the benefits and drawbacks of a pure appointment and a pure walk-in policy. The optimal policy successfully counterbalances the non-stationary nature of walk-in arrivals at both the daily and weekly levels, by prescribing how many appointment slots to reserve and at which times.

Kronecker product covariance structure models

Beata Ros (Vrije Universiteit Amsterdam) Waaier 2, vr 12:32-12:35

We investigate properties of the Kronecker product covariance structure models. Namely, we assume that is a random matrix and vec (X) ~ with some additional assumptions about matrices and . A nice property is that the matrix has much fewer parameters than the unrestricted covariance matrix .

Suppose we have X1, . . . ,XN -the data. We are interested in estimating with use of the data. We consider maximum likelihood estimation, thus obtain maximum likelihood equations. We investigate two aspects of these equations. First is the existence of solutions of the equations. The second is uniqueness of the solution. Depending on additional assumptions about and as well as p, q -number of rows, columns of X and N -number of samples, we have different properties of the likelihood equations with respect to existence and uniqueness of solutions. We are interested in using this model for the analysis of simultaneously collected EEG and fMRI data.

Mathematics teachers’ professional development by means of lesson study

Daan van Smaalen (Universiteit Twente) Waaier 2, vr 12:35-12:38

Lesson study is a professional development practice in which teachers collaborate to develop a lesson, teach and observe the lesson to collect data on student learning and development, and use their observations to refine their lesson. It is a process that teachers engage in to learn more about teaching and learning, it is not about designing a perfect lesson. Observing students and discussing the lesson and instruction more broadly are central activities of lesson study.

Lesson study originated in Japan, where this practice is seen as the means for teacher development for several decades now. Since the millennium lesson study is slowly receiving some attention in western countries like America, Britain and Australia. In the Netherlands, the first initiatives are launched recently. The phenomenon lesson study is thus hardly documented. Therefore, there is a need to expand the knowledgebase regarding this professional development practice. The central research question related to my PhD research is: What and how do teachers learn when they participate in a lesson study team?

Planning and scheduling of semi-urgent surgeries

Maartje E. Zonderland (Universiteit Twente / Leids Universitair Medisch Centrum); Waaier 2, vr 12:39-12:42

We consider a surgical department where elective, semi-urgent and urgent patients patients are treated. The latter patient type needs treatment immediately, which is carried out in a separate emergency OR; semi-urgent patients, who arrive unexpectedly, need surgery within one or two weeks and are treated, just as elective patients, in regular OR time.

Elective patients are canceled to accommodate semi-urgent patients, which is highly undesirable from a patient perspective. Therefore a part of regular OR capacity is dedicated to semi-urgent patients. However, no semi-urgent patients may show up. Since elective patients cannot be planned on such short notice, scarce OR time is not used, which is very undesirable from a financial point of view. We describe a methodology, based on a queueing theory approach, to handle the uncertainty caused by semi-urgent patients.

Decentralized Scheduling on Related Machines

Ruben Hoeksma (Universiteit Twente) Waaier 2, vr 12:42-12:45

The price of anarchy measures by how much the performance of a system deteriorates due to the lack of central coordination. We address the classical uniformly related machine scheduling problem. This problem treats the scheduling of multiple jobs with different lengths on a number of parallel machines with different speeds. We assume that the jobs may choose the machine on which they are processed. When jobs seek to minimize their own completion time, the utilitarian social choice function is to minimize the average job completion time. In this setting we analyze the price of anarchy for the natural coordination mechanism where jobs are sequenced shortest first per machine. We show that the price of anarchy is bounded from below by 1.58 and from above by 2. This complements recent results on the price of anarchy for the more general unrelated machine scheduling problem by Cole et al. Moreover, as Nash equilibria correspond one-to-one to SPT schedules, the same bounds hold for the SPT heuristic on uniformly related machines. Thereby, our work also fills a gap in the literature.