The power to mass ratio of a heavy truck causes even moderate slopes to have a significant influence on the motion. The velocity will inevitable vary within an interval that is primarily determined by the ratio and the road topography. If further variations are actuated by a controller, there is a potential to lower the fuel consumption by taking the upcoming topography into account. This possibility is explored through theoretical and simulation studies as well as experiments in this work. Look-ahead control is a predictive strategy that repeatedly solves an optimization problem online by means of a tailored dynamic programming algorithm. The scenario in this work is a drive mission for a heavy diesel truck where the route is known. It is assumed that there is road data on-board and that the current heading is known. A look-ahead controller is then developed to minimize fuel consumption and trip time. The look-ahead control is realized and evaluated in a demonstrator vehicle and further studied in simulations. In the prototype demonstration, information about the road slope ahead is extracted from an on-board database in combination with a GPS unit. The algorithm calculates the optimal velocity trajectory online and feeds the conventional cruise controller with new set points. The results from the experiments and simulations confirm that look-ahead control reduces the fuel consumption without increasing the travel time. Also, the number of gear shifts is reduced. Drivers and passengers that have participated in tests and demonstrations have perceived the vehicle behavior as comfortable and natural. / <p>Report code: LIU-TEK-LIC-2007:28.</p>

Many tasks of visual computing and communications such as object recognition, matting, compression, etc., need to extract and encode the outer boundary of the object in a digital image or video. In this thesis, we focus on a particular video segmentation task and propose an efficient method for head-and-shoulder of humans through video frames. The key innovations for our work are as follows: (1) a novel head descriptor in polar coordinate is proposed, which can characterize intrinsic head object well and make it easy for computer to process, classify and recognize. (2) a learning-based method is proposed to provide highly precise and robust head-and-shoulder segmentation results in applications where the head-and-shoulder object in the question is a known prior and the background is too complex. The efficacy of our method is demonstrated on a number of challenging experiments. / Thesis / Master of Applied Science (MASc)

No description available.

This thesis examines decision optimization of product recalls. Product recalls in recent years have shown unprecedented impact on both immediate economic and reputational damage to the company and long-lasting impact on the brand and industry. Admittedly, imperfect product quality makes recalls inevitable. Thus, we explore from three perspectives to elicit business insights regarding better management and risk control. Chapter 1 introduces the topic of product recall management optimization and its real-world motivation. Chapter 2 views the decision making of "when to initiate a product recall" as a dynamic process and takes the feedback of customer returns to update the product defect rate. Updating is simplified by the conjugate properties of beta distribution and Bernoulli trials. We develop the optimal stopping model to find the thresholds of total product returns above which initiating recall is optimal. We implement dynamic programming to solve the model optimally. For large-size problems, we propose a simulation method to balance computation time with solution quality. Chapter 3 allows the company to control the recall risk by investing in quality. We adopt the one-stage stochastic newsvendor model and add quality-dependent recall risk. The resulting model is not concave in production quantity and quality levels. The parametric analysis reveals several interesting features such as the optimal ordering quantity and quality level have a conflicting relationship. We further extend our model from internal supply to external supply from multiple sources. Chapter 4 examines managing product recalls from the closed-loop supply chain management and disruption management perspectives. We model the location and allocation decisions of both manufacturing plants and reprocessing facilities where facilities are built after the recalls. Numerical experiments show the costs of overlooking potential recalls vary greatly, indicating the necessity of considering recalls in initial designs and the importance of accurate recall probability prediction. Chapter 5 summarizes. / Thesis / Doctor of Philosophy (PhD)

The future success of chargeable vehicles will, among other factors, depend on their charging costs and their ability to charge with minimal disturbances to the national, local and household electrical grid. To be able to minimize costs and schedule charging sessions, there has to be knowledge of how the charging power varies with time. This is called charging profile. A number of charging profiles for a Volvo V60 plug‑in hybrid electric vehicle have been recorded. For charging currents above 10 A they prove to be more complex than are assumed in most current research papers.   The charging profiles are used together with historical electricity prices to calculate charging costs for 2013 and 2014. Charging is assumed to take place during the night, between 18:00 and 07:00, with the battery being totally depleted at 18:00. By using a timer to have the charging start at 01:00, instead of immediately at 18:00, annual charging costs are reduced by approximately 7 to 8%. By using dynamic programming to optimize the charging sessions, annual charging costs are reduced by approximately 10 to 11%. An interesting issue regarding dynamic programming was identified, namely when using a limited set of predetermined discrete control signals, interpolation returns unrealizable cost-to-go values. This occurs specifically for instances crossing the zero cost-to-go area boundary.   It is concluded that the mentioned savings are realizable, via implementing timers or optimization algorithms into consumer charging stations. Finally, by using these decentralized charging planning tools and seen from a power usage perspective, at least 30% of the Swedish vehicle fleet could be chargeable and powered by the electrical grid.

published_or_final_version / Mathematics / Master / Master of Philosophy

摘要 本文於確定給付退休金計劃下，探討基金經理人於最差基金財務短絀情境發生前極大化管理目標之最適投資組合，基金比值過程定義為基金現值與負債指標之比例，管理人將於指定最差基金比值發生前極大化達成既定經營目標之機率，隨時間改變之基金投資集合包括無風險之現金、債券與股票。本研究建構隨機控制模型描述此最適化問題，並以動態規劃方法求解，由結果歸納，經理人之最適策略包含極小化基金比值變異之避險因素，風險偏好及跨期投資集合相關之避險因素與模型狀態變數相關之避險因素。本研究利用馬可夫練逼近法逼近隨機控制的數值解，結果顯示基金經理人須握有很大部位的債券，且不同的投資期間對於最適投資決策有很大的影響。 關鍵字: 短絀、確定給付、負債指標、隨機控制、動態規劃。 / Abstract This paper analyzes the portfolio problem that is a pension fund manager has to maximize the possibility of reaching his managerial goal before the worst scenario shortfall occurs in a defined benefit pension scheme. The fund ratio process defined as the ratio between the fund level and its accrued liability benchmark is attained to maximize the probability that the predetermined target is achieved before it falls below an intolerable boundary. The time-varying opportunity set in our study includes risk-free cash, bonds and stock index. The problems are formulated as a stochastic control framework and are solved through dynamics programming. In this study, the optimal portfolio are characterized by three components, the liability hedging component, the intertemporal hedging component against changes in the opportunity set, and the temporal hedging component minimizing the variation in fund ratio growth. The Markov chain approximation methods are employed to approximate the stochastic control solutions numerically. The result shows that fund managers should hold large proportions of bonds and time horizon plays a crucial role in constructing the optimal portfolio. Keywords: shortfall; defined benefit; liability benchmark; stochastic control; dynamic programming.

This research is concerned with scheduling production over a finite planning horizon in a capacitated manufacturing facility. It is assumed that a second source of supply is available by means of subcontracting and that the demand varies over time. The problem is to establish the production level in the facility and/or the ordering quantity from the subcontractor for each period in the planning horizon. Firstly, the cost functions are analyzed and two types of realistic production cost models are identified. Then mathematical models are developed for two different problems. One is a single criterion problem aimed at minimizing the total production and inventory costs. The other is a bicriterion problem which seeks the efficient frontier with respect to the total cost and the number of subcontractings, both to be minimized, over the planning horizon. For each of the above, two methods, namely, a general dynamic programming approach and an improved dynamic programming approach (Shortest path method) are presented. Several results are obtained for reducing the computations in solving these problems. Based on these results, algorithms are developed for both problems. The computational complexity of these algorithms are also analyzed. Two heuristic rules are then suggested for obtaining near-optimal solutions to the first problem with lesser computation. Both rules have been tested extensively and the results indicate advantages of using them. One of these rules is useful for solving the uncapacitated problem faster without losing optimality. The above results are then extended to other cases where some of the assumptions in the original problem are relaxed. Finally, we studied the multi-item lot-sizing problem with the subcontracting option and proposed a heuristic for solving the problem by the Lagrangean relaxation approach. We demonstrated that with an additional capacity constraint in the dual problem the feasible solution and the lower bound obtained during each iteration converge much faster than without it. After testing some randomly generated problems we found that most of the solutions obtained from the heuristic are very close to the best lower bound obtained from the dual problem within a limited number of iterations.

A stochastic dynamic programming model (called P.B.S.D.P.) based on the consideration of peak discharge and time between peaks as two stochastic variables has been used to model and to solve a reservoir operation problem. This conceptualization of the physical reality allows to solve, in this order, the tactical and strategic operation of surface reservoirs. This P.B.S.D.P. model has been applied to the Sau reservoir in the Northeastern corner of Spain. The results showed a significant improvement over the currently used operation procedure, yielding values of yearly average electricity production that are somewhat under 6% of what could have been the maximum electricity production.

The areas of machine learning, data mining, and knowledge representation have many different formats used to represent information. Decision rules, amongst these formats, are the most expressive and easily-understood by humans. In this thesis, we use dynamic programming to design decision rules and analyze them. The use of dynamic programming allows us to work with decision rules in ways that were previously only possible for brute force methods. Our algorithms allow us to describe the set of all rules for a given decision table. Further, we can perform multi-stage optimization by repeatedly reducing this set to only contain rules that are optimal with respect to selected criteria. One way that we apply this study is to generate small systems with short rules by simulating a greedy algorithm for the set cover problem. We also compare maximum path lengths (depth) of deterministic and non-deterministic decision trees (a non-deterministic decision tree is effectively a complete system of decision rules) with regards to Boolean functions. Another area of advancement is the presentation of algorithms for constructing Pareto optimal points for rules and rule systems. This allows us to study the existence of “totally optimal” decision rules (rules that are simultaneously optimal with regards to multiple criteria). We also utilize Pareto optimal points to compare and rate greedy heuristics with regards to two criteria at once. Another application of Pareto optimal points is the study of trade-offs between cost and uncertainty which allows us to find reasonable systems of decision rules that strike a balance between length and accuracy.