Organizational Unit: Department of Industrial Engineering
Loading...
Date established
City
Country
ID
66 results
Publication Search Results
Now showing 1 - 10 of 66
Master ThesisPublication Metadata only Thermal request optimization of a smart district heating systemKarasu, Mehmet Berk; Yanıkoğlu, İhsan; Yanıkoğlu, İhsan; Önal, Mehmet; Yavuz, T.; Department of Industrial EngineeringThis thesis proposes a solution approach to manage the heating plans of tenants served by a district heating plant located in Sweden to decrease the carbon footprint of the residents. To do that, the daily temperature request of each household in the associated pilot region is obtained, and the daily temperature profile of each household is optimized with the help of a decision support system and smart valves. The hot water inflow rates of radiators are remotely controlled via smart valves at each flat to minimize the total energy consumption, carbon emission and cost associated with the energy consumption of the district heating plant. We aim to shave the peak demands while fully satisfying the temperature requests of households without violating their thermal comfort. Peak demand shaving is achieved by generating preheating schedules via mathematical optimization and using the thermal storage potential of the insulated flats. The resulting mathematical optimization model presents significant computational challenges that cannot be efficiently solved using optimization solvers within a reasonable time limit. To this end, we adapt three genetic algorithm approaches that are computationally scalable for realistically-sized instances and verified to yield near-optimal solutions for the test instances. Extensive numerical analyses how the effectiveness of the proposed approach and the genetic algorithm since we yield significant carbon emission decrease and cost savings compared with the method that the experts of the utility company propose.Master ThesisPublication Metadata only Parametric and non parametric models for stochastic next day operating room scheduling(2018-01) Sevindik, Ömer Hikmet; Kayış, Enis; Kayış, Enis; Güler, M. Güray; Çoban, Elvin; Department of Industrial Engineering; Sevindik, Ömer HikmetOperating rooms are the resources that generate the most part of the revenue of hospitals. On the other hand, they generate the most part of the expenses, as well. Because of the uncertainty of surgery durations, scheduling operating rooms are very difficult. But their impact on the finances of a hospital makes it vital for the planners to carry out scheduling as best as they can. Another problem that lies in the way of fine operating room scheduling is limited surgery data available for use. Uncertainty and diversity of surgeries that may take place in a given operating room makes it difficult to obtain sufficient amount of surgery duration data. In this study we describe a stochastic optimization model for computing OR schedules that are effected by the uncertainty in surgery durations. We focus on scheduling start times. We show that our model can be used to generate substantial reductions in OR team waiting, OR idling, overtime costs. The model in this study is studied with 3 solution approaches: (i) parametric approach, (ii) non parametric approach, (iii) a simple but practical heuristic. Considering all scenarios in this study, parametric approach manages to perform 6,18% close to optimal solution, whereas non parametric approach performs 7,66% and heuristic approach performs 78,17% close to optimal solution. When compared to non parametric approach, parametric approach performs better when number of historical surgery duration sample size is small. In contrast, when the number of historical surgery duration sample size is large, non parametric approach starts performing better. All three solution approaches provide meaningful results, where parametric approach performs better in most cases when compared to other solution approaches.Master ThesisPublication Metadata only A robust longitudinal model for song popularity: a cross-cultural study(2021-08-16) Çimen, Ahmet Can; Kayış, Enis; Kayış, Enis; Ulukuş, M. Y.; Kaya, M.; Department of Industrial Engineering; Çimen, Ahmet CanUsage of new generation music streaming platforms such as Spotify and Apple Music has increased rapidly in the last years. Understanding the music preferences of the user base is valuable for these firms, translating into higher customer satisfaction. In this study, we develop and compare several statistical models to quantify the effects of the different factors on music popularity by using acoustic and artist-related features. We compare results from three countries to understand whether there are any cultural differences in how much each of these factors affects song popularity. To compare the results, we use weekly top 200 charts and songs’ acoustic features as data sources. In addition to acoustic features, we add acoustic similarity, genre, song recentness features into the dataset. We apply the Flexible Least Squares (FLS) method and developed Optimal Stepwise Linear Regression (SLR) methods to and observe time-varying regression coefficients. We also propose a regression tree-based heuristic algorithm to solve the SLR problem in a reasonable time. The FLS method tries to keep the differences of adjacent weeks’ coefficients as small as possible. On the other hand, the SLR does not control the variation between consecutive weeks’ coefficients however, it only allows a limited number of coefficient changes over time. Coefficients that we obtain from the FLS method show that we can keep track of the changes of the factors that affect music listening habits and associate real-life events with these changes. Besides, we propose for the SLR method, we can quickly achieve near-optimal solutions which give us clues about the time of the significant changes in the music taste in different countries. Finally, inferences we made with the study may help the music industry grow and contribute to the anthropological fields for discovering socio-cultural/political aspects of the music taste changes.Master ThesisPublication Metadata only An integrated post-disaster assessment routing problem for collecting damage information with dronesAdsanver, Birce; Göktürk, Elvin Çoban; Göktürk, Elvin Çoban; Koyuncu, Burcu Balçık; Albey, Erinç; Yanıkoğlu, İhsan; Yıldırım, U. M.; Department of Industrial Engineering; Adsanver, BirceIn this study, we focus on post-disaster damage assessment operations supported by a set of drones when the disaster-affected area is divided into grids, and grids are clustered based on their attributes. We propose a two-phase methodology to assess the damage status of the built environment in grids. Specifically, given a set of drones and a limited time for an assessment interval, the first phase addresses the problem of determining the grids to scan by each drone and the sequence of visits to the selected grids. We aim to maximize the total priority score collected from the scanned grids while satisfying the predefined targeted coverage ratio. In the second phase, we aim to predict the damage status of unscanned grids by using the cluster-based information obtained from the scanned grids at the end of the assessment interval. Nevertheless, the damage status of all grids may not be assessed (by scanning or prediction) after one interval; therefore, these two phases iterate until all grids are evaluated. For the problem solved in the first phase, we adapt two formulations from the literature developed for electric vehicle routing problems. We also develop a Variable Neighborhood Descent based heuristic which can find high-quality solutions rapidly. We evaluate the performance of the alternative formulations and the heuristic in a variety of instances. For the second phase, we devise a novel imputation method and different imputation policies to predict the damage status of the unscanned grids. We also define several performance metrics to measure the efficiency and effectiveness of the proposed imputation policies. Our analyses demonstrate that using the proposed imputation policies improve the system performance as they induce a rapid detection of the damaged areas.Master ThesisPublication Metadata only An application of unrelated parallel machine scheduling with sequence-dependent setups at Vestel Electronic(2017-08) Sarıkaya, Merve Burcu; Özener, Okan Örsan; Özener, Okan Örsan; Ekici, Ali; Duran, S.; Department of Industrial Engineering; Sarıkaya, Merve BurcuVestel Electronics produces LCD/LED televisions and has a significant market share in consumer electronics sector in Europe. TV manufacturing is planned based on a make-to order strategy, and Vestel uses 15 assembly lines to produce 110 different product groups and 3817 different models. Once the orders are received, production scheduling is performed at the beginning of each month, and the goal is to satisfy the demand on time as much as possible. Each order/job is processed on one of the compatible assembly lines, and preemption is not allowed. In this thesis, we study the TV production scheduling operations at Vestel. The problem faced by Vestel is a variant of unrelated machine scheduling problem, and the objective is to minimize total tardiness. We propose a wide range of heuristics including a very simple sequential algorithm and a novel set partitioning-based approach. We test the heuristics on the real-life data and compare the solutions with the current practice. We observe up to 50% improvement in total tardiness. Keywords: parallel machine scheduling; unrelated machines; sequence-dependent setups; tardinessMaster ThesisPublication Metadata only Coordinated inventory planning for humanitarian relief agencies(2016-07) Karaca, Meserret; Özener, Okan Örsan; Koyuncu, Burcu Balçık; Özener, Okan Örsan; Koyuncu, Burcu Balçık; Özen, Ulaş; Ekici, Ali; Duran, S.; Department of Industrial Engineering; Karaca, MeserretPre-positioning relief supplies in strategic locations around the world is essential for effective disaster response, especially during the critical 72 hours immediately following the disaster. Most of the existing studies that use quantitative models to determine pre-positioning decisions focus on a single relief agency and assume that the agency makes stock pre-positioning decisions independently of other agencies; that is, the possibility of sharing inventory among different agencies is not considered. In this study, we aim to investigate the potential benefits of making stock pre-positioning decisions collaboratively among multiple agencies. In particular, we consider two agencies that stock relief supplies in a joint depot owned and operated by a separate coordinator (such as the United Nations Humanitarian Response Depot). We assume that these agencies have several operating regions throughout the world. Each operating region may be served by a single agency or by several different agencies depending on its location. Once a disaster occurs in an agency's operating region, the agency aims to satisfy demand as much as possible. Also, other agencies may share their excess inventory with the responding agency. The amount of supplies that can be sent to the disaster region is affected by the uncertain post-disaster funding level of the responding agency. We consider a finite set of scenarios to characterize the uncertainties in disaster locations, impacts and post-disaster funding levels and develop a two-stage stochastic programming model to determine the amount of inventory to be pre-positioned at the joint depot by each agency. We perform a numerical analysis to establish when collaborative action would be beneficial for different types of agencies in different settings.Master ThesisPublication Metadata only Two-level influence maximization problem under deterministic linear threshold modelEşki, Doruk; Danış, Dilek Günneç; Danış, Dilek Günneç; Albey, Erinç; Güney, E.; Department of Industrial Engineering; Eşki, DorukData-driven decision-making strategies can make online marketing more efficient and help companies reach a larger number of customers using limited resources. In this respect, the Influence Maximization Problem searches for a certain number of influential individuals on a social network so that the information/product spread initiated from such individuals is maximized. We introduce a novel problem, the Two-Level Influence Maximization Problem, which allows influencing neighbors without eventually adopting the product. To solve this problem, we develop a Greedy Algorithm and two variants, namely Modified Greedy Algorithm and Update Limited Algorithm. Modified Greedy Algorithm reaches the same objective function value in a much lower runtime. In Update Limited Greedy Algorithm, runtime is further reduced but objective function value may be compromised. We also introduce a Simulated Annealing-based metaheuristic with a tabu strategy that exploits the problem-specific neighborhood moves. We apply the Simulated Annealing algorithm with the proposed Greedy Algorithm and its variants on small scale random networks and with PageRank and HITS algorithms as initial solutions on large scale real social networks. We also show that following alternative strategies such as maximizing number of individuals that influence their neighbors without adopting the product can also return an effective starting solution for Simulated Annealing algorithm. Computational experiments on simulated and real social networks show that our heuristics can provide high-quality solutions.Master ThesisPublication Metadata only Reducing open vial wastage with lateral transshipment in COVID-19 vaccine administrationÖzmemiş, Çağrı; Koyuncu, Burcu Balçık; Koyuncu, Burcu Balçık; Yanıkoğlu, İhsan; Erçetin, E. D. G.; Department of Industrial Engineering; Özmemiş, ÇağrıAfter the discovery and production of vaccines for the immunization against COVID-19 pandemic, mass vaccination programs are launched across the world. One of the main challenges faced during the COVID-19 mass vaccination programs is vaccine wastage. In a preliminary and empirical work conducted with healthcare professionals working at family health centers, we identify that using multi-dose vaccine vials is one of the significant factors causing the wastage. Some COVID-19 vaccines are preserved in multi-dose vials, and once a vial is opened, doses inside perish and go to waste in a few hours if not used. In this study, we analyze the lateral transshipment strategies in which open vials are transferred between vaccination points to decrease vaccine wastage. Besides making operational decisions regarding the vaccination process, we also examine the demand uncertainty issue due to the appointment no-shows. The problem is formulated as a 2-stage stochastic programming model. We solve the problem by developing a heuristic approach in which we cluster the vaccination points to form alliances using a mathematical model and utilizing the second stage of the 2-stage stochastic programming model. We validate our work by conducting a case study with family health centers in Tuzla, İstanbul and provide valuable insights for the case. The results show that our heuristic approach can be relied on its reasonable solution time, and provide quality solutions.Master ThesisPublication Metadata only A syncronized routing problem for restoring inter-dependent infrastructure networks(2018-05) Sevindik, Büşra; Günneç, Dilek; Balçık, Burcu; Günneç, Dilek; Balçık, Burcu; Ekici, Ali; Yanıkoğlu, İhsan; Çavdaroğlu, B.; Department of Industrial Engineering; Sevindik, BüşraDisasters may cause significant damages in lifeline infrastructure systems (such as gas, power, water) and lead to long-lasting failures. It is important to repair the damaged components and restore the affected infrastructures quickly. Since different lifeline infrastructure systems depend on each other, considering the inter-dependencies among different networks during repair planning can speed up the recovery process. In this thesis, we focus on developing practical methods to support planning repair operations for two inter-dependent infrastructure networks by considering the inter-dependencies within and between these networks. Specifically, we assume that repairing a damaged component may not be sufficient for making the component functional due to network dependencies. We consider multiple repair teams, each of which can repair the damaged components of one type of infrastructure, and formulate a coordinated repair scheduling problem, which determines a repair schedule for each repair team to minimize the total time for making all nodes functional. To solve this problem, we propose two alternative constructive heuristics, which employ different strategies to prioritize the visit of the damaged nodes based on their dependency status. We also apply local search procedures to improve the solutions attained by the constructive heuristics. We present computational results to evaluate the performance of the proposed heuristics. The results show that our heuristics lead to high quality solutions and can be used to make repair plans quickly in the post-disaster environment.Master ThesisPublication Metadata only An adaptive large neighborhood search for the multi-compartment inventory routing problem(2021-06-10) Gültekin, Ceren; Özener, Okan Örsan; Özener, Okan Örsan; Yanıkoğlu, İhsan; Ekici, Ali; Önal, Mehmet; Yakıcı, E.; Department of Industrial Engineering; Gültekin, CerenIn this thesis study, we concentrate on an inventory routing problem with a fleet of multi-compartment vehicles which enables the distribution of different products to customers on a delivery route. Using separate compartments on a vehicle increases profitability and customer satisfaction when customer demands vary over product and period basis. We assume that the compartment that each product can be loaded is known and the capacities of the compartments are fixed. Customers have preset storage capacities and distribution plans should be made in a way that no customers would face stock-outs for any product on any day. We observe the practices of this variant in the distribution of foods with different temperature needs to groceries, feed distribution to livestock farms, and collection of different types of recyclable wastes. We examine this problem separately for three assumptions considering different cases of allowing/disallowing split delivery to customers. We propose a matheuristic ap proach to solve the addressed problem where we systematically integrate an Adaptive Large Neighborhood Search algorithm with mathematical programming models. We generate a set of instances and test the performance of our algorithm by comparing it with the results obtained by a flow formulation adapted from the literature. We observe that the best results we find for each instance are only %11.7 worse than the solutions found by the flow formulation on average.