Uncertainty surrounds the future evolution of key factors affecting the attractiveness of various nuclear fuel cycles, rendering the concept of a unique optimal fuel cycle transition strategy invalid. This work applies decision-making under uncertainty to fuel cycle transition analysis, demonstrating a new, systematic methodology for choosing flexible, adaptable hedging strategies that yield middle-of-the-road results until uncertainties are resolved. A case study involving transition from the current once-through light water reactor (LWR) fuel cycle to one relying on continuous recycle in fast reactors (FRs) is cast as a no-data decision problem. The transition is subject to uncertainty in the cost of spent nuclear fuel (SNF) and high-level waste (HLW) disposal in a geologic repository, slated to open some years into the future. Following the repository open date, the cost of SNF and HLW disposal is made known, and may take on one of five possible values. Strategies for the transition are enumerated and simulated using VEGAS, a systems model of the nuclear fuel cycle that solves for its material balance and applies input cost data to calculate the associated annual levelized cost of electricity (LCOE). Perfect information strategies are found using the lowest average, maximum, and integrated LCOE objective functions. The loss in savings for following a strategy other than the perfect information strategy is the “regret” which is calculated by evaluating the performance of each strategy for every end-state. Hedging strategies are then selected by either minimizing the maximum or the expected regret. Generally, the optimal hedging strategy identified using the decision methodology suggests a partial transition to a closed fuel cycle prior to the repository open date. Once the repository opens, the transition may be abandoned or accelerated depending on which disposal cost outcome is realized. The lowest average and integrated LCOE objective functions perform similarly; however, the lowest maximum LCOE objective function appears overly sensitive to aberrations in the annual LCOE that arise due to idle reprocessing capacity. The minimax regret choice criterion is shown to be more conservative than the lowest expected regret choice criterion, as it acts to hedge against the worst-case outcome. By following a hedging strategy, agents may alter their fuel cycle strategy more readily once uncertainties are resolved. This results since hedging strategies provide flexibility in the nuclear fuel cycle, preserving what options exist. To this end, the work presented here may provide guidance for agent-based, behavioral modeling in fuel cycle simulators, as well as decision-making in real world applications. / text

This dissertation presents efficient, on-line, convergent methods to find defense strategies against attacks in dynamic multi-stage attacker-defender games including adaptive learning. This effort culminated in four papers submitted to high quality journals and a book and they are partially published. The first paper presents a novel fictitious play approach to describe the interactions between the attackers and network administrator along a dynamic game. Multi-objective optimization methodology is used to predict the attacker's best actions at each decision node. The administrator also keeps track of the attacker's actions and updates his knowledge on the attacker's behavior and objectives after each detected attack, and uses this information to update the prediction of the attacker's future actions to find its best response strategies. The second paper proposes a Dynamic game tree based Fictitious Play (DFP) approach to describe the repeated interactive decision processes of the players. Each player considers all possibilities in future interactions with their uncertainties, which are based on learning the opponent's decision process (including risk attitude, objectives). Instead of searching the entire game tree, appropriate future time horizons are dynamically selected for both players. The administrator keeps tracking the opponent's actions, predicts the probabilities of future possible attacks, and then chooses its best moves. The third paper introduces an optimization model to maximize the deterministic equivalent of the random payoff function of a computer network administrator in defending the system against random attacks. By introducing new variables the transformed objective function becomes concave. A special optimization algorithm is developed which requires the computation of the unique solution of a single variable monotonic equation. The fourth paper, which is an invited book chapter, proposes a discrete-time stochastic control model to capture the process of finding the best current move of the defender. The defender's payoffs at each stage of the game depend on the attacker's and the defender's accumulative efforts and are considered random variables due to their uncertainty. Their certain equivalents can be approximated based on their first and second moments which is chosen as the cost functions of the dynamic system. An on-line, convergent, Scenarios based Proactive Defense (SPD) algorithm is developed based on Differential Dynamic Programming (DDP) to solve the associated optimal control problem.

The development of suitable long term flood risk intervention strategies is a challenge. Climate change alone is a significant complication but in addition complexities exist trying to identify the most appropriate set of interventions, the area with the highest economical benefit and the most opportune time for implementation. All of these elements pose difficulties to decision makers. Recently, there has been a shift in the current practice for appraising potential strategies and consideration is now being given to ensure flexible, adaptive strategies to account for the uncertain climatic conditions. Real Options in particular is becoming an acknowledged approach to account for the future uncertainties inherent in a flood risk investment decision. Real Options facilitates adaptive strategies as it enables the value of flexibility to be explicitly included within the decision making process. Opportunities are provided for the decision maker to modify and update investments when knowledge of the future state comes to light. In this thesis the use of Real Options in flood risk management is investigated as a method to account for the uncertainties of climate change. Each Intervention strategy is purposely designed to capture a level of flexibility and have the ability to adapt in the future if required. A state of the art flood risk analysis tool is employed to evaluate the risk associated to each strategy over future points in time. In addition to Real Options, this thesis also explores the use of evolutionary optimisation algorithms to aid the decision making process when identifying the most appropriate long term strategies. Although the risk analysis tool is capable of quantifying the potential benefits attributed to a strategy, it is not necessarily able to identify the most appropriate. Methods are required which can search for the optimal solutions according to a range of performance metrics. Single and multi-objective genetic algorithms are investigated in this thesis as a method to search for the most appropriate long term intervention strategies. The Real Options concepts are combined with the evolutionary multiobjective optimisation algorithm to create a decision support methodology which is capable of searching for the most appropriate long term economical yet robust intervention strategies which are flexible to future change. The methodology is applied to two individual case studies, a section of the Thames Estuary and an area on the River Dodder. The results show the inclusion of flexibility is advantageous while the outputs provide decision makers with supplementary knowledge which previously has not been considered.

Increasing the percentage of wind power in the United States electricity generation mix would facilitate the transition towards a more sustainable, low-pollution, and environmentally-conscious electricity grid. However, this effort is not without cost. Wind power generation is time-variable and typically not synchronized with electricity demand (i.e., load). In addition, the highest-output wind resources are often located in remote locations, necessitating transmission investment between generation sites and load. Furthermore, negative public perceptions of wind projects could prevent widespread wind development, especially for projects close to densely-populated communities. The work presented in my dissertation seeks to understand where it’s best to locate wind energy projects while considering these various factors. First, in Chapter 2, I examine whether energy storage technologies, such as grid-scale batteries, could help reduce the transmission upgrade costs incurred when siting wind projects in distant locations. For a case study of a hypothetical 200 MW wind project in North Dakota that delivers power to Illinois, I present an optimization model that estimates the optimal size of transmission and energy storage capacity that yields the lowest average cost of generation and transmission ($/MWh). I find that for this application of storage to be economical, energy storage costs would have to be $100/kWh or lower, which is well below current costs for available technologies. I conclude that there are likely better ways to use energy storage than for accessing distant wind projects. Following from this work, in Chapter 3, I present an optimization model to estimate the economics of accessing high quality wind resources in remote areas to comply with renewable energy policy targets. I include temporal aspects of wind power (variability costs and correlation to market prices) as well as total wind power produced from different farms. I assess the goal of providing 40 TWh of new wind generation in the Midwestern transmission system (MISO) while minimizing system costs. Results show that building wind farms in North/South Dakota (windiest states) compared to Illinois (less windy, but close to population centers) would only be economical if the incremental transmission costs to access them were below $360/kW of wind capacity (break-even value). Historically, the incremental transmission costs for wind development in North/South Dakota compared to in Illinois are about twice this value. However, the break-even incremental transmission cost for wind farms in Minnesota/Iowa (also windy states) is $250/kW, which is consistent with historical costs. I conclude that for the case in MISO, building wind projects in more distant locations (i.e., Minnesota/Iowa) is most economical. My two final chapters use semi-structured interviews (Chapter 4) and conjoint-based surveys (Chapter 5) to understand public perceptions and preferences for different wind project siting characteristics such as the distance between the project and a person’s home (i.e., “not-in-my-backyard” or NIMBY) and offshore vs. onshore locations. The semi-structured interviews, conducted with members of a community in Massachusetts, revealed that economic benefit to the community is the most important factor driving perceptions about projects, along with aesthetics, noise impacts, environmental benefits, hazard to wildlife, and safety concerns. In Chapter 5, I show the results from the conjoint survey. The study’s sample included participants from a coastal community in Massachusetts and a U.S.-wide sample from Amazon’s Mechanical Turk. Results show that participants in the U.S.-wide sample perceived a small reduction in utility, equivalent to $1 per month, for living within 1 mile of a project. Surprisingly, I find no evidence of this effect for participants in the coastal community. The most important characteristic to both samples was the economic benefits from the project – both to their community through increased tax revenue, and to individuals through reduced monthly energy bills. Further, participants in both samples preferred onshore to offshore projects, but that preference was much stronger in the coastal community. I also find that participants from the coastal community preferred expanding an existing wind projects rather than building an entirely new one, whereas those in the U.S.-wide sample were indifferent, and equally supportive of the two options. These differences are likely driven by the prior positive experience the coastal community has had with an existing onshore wind project as well as their strong cultural identity that favors ocean views. I conclude that preference for increased distance from a wind project (NIMBY) is likely small or non-existent and that offshore wind projects within 5 miles from shore could cause large welfare losses to coastal communities. Finally, in Chapter 6, I provide a discussion and policy recommendations from my work. Importantly, I recommend that future research should combine the various topics throughout my chapters (i.e., transmission requirements, hourly power production, variability impacts to the grid, and public preferences) into a comprehensive model that identifies optimal locations for wind projects across the United States.

This thesis investigates the following question: Can supervised learning techniques be successfully used for finding better solutions to multistage stochastic programs? A similar question had already been posed in the context of reinforcement learning, and had led to algorithmic and conceptual advances in the field of approximate value function methods over the years. This thesis identifies several ways to exploit the combination "multistage stochastic programming/supervised learning" for sequential decision making under uncertainty. Multistage stochastic programming is essentially the extension of stochastic programming to several recourse stages. After an introduction to multistage stochastic programming and a summary of existing approximation approaches based on scenario trees, this thesis mainly focusses on the use of supervised learning for building decision policies from scenario-tree approximations. Two ways of exploiting learned policies in the context of the practical issues posed by the multistage stochastic programming framework are explored: the fast evaluation of performance guarantees for a given approximation, and the selection of good scenario trees. The computational efficiency of the approach allows novel investigations relative to the construction of scenario trees, from which novel insights, solution approaches and algorithms are derived. For instance, we generate and select scenario trees with random branching structures for problems over large planning horizons. Our experiments on the empirical performances of learned policies, compared to golden-standard policies, suggest that the combination of stochastic programming and machine learning techniques could also constitute a method per se for sequential decision making under uncertainty, inasmuch as learned policies are simple to use, and come with performance guarantees that can actually be quite good. Finally, limitations of approaches that build an explicit model to represent an optimal solution mapping are studied in a simple parametric programming setting, and various insights regarding this issue are obtained.

Growing attention is being paid to the problem of efficiently designing and operating reverse supply chain systems to handle the return flows of production wastes, packaging, and end-of-life products. Because uncertainty plays a significant role in all fields of decision-making, solution methodologies for determining the strategic infrastructure of reverse production systems under uncertainty are required. This dissertation presents innovative optimization algorithms for designing a robust network infrastructure when uncertainty affects the outcomes of the decisions. In our context, robustness is defined as minimizing the maximum regret under all realization of the uncertain parameters. These new algorithms can be effectively used in designing supply chain network infrastructure when the joint probability distributions of key parameters are unknown. These algorithms only require the information on potential ranges and possible discrete values of uncertain parameters, which often are available in practice. These algorithms extend the state of the art in robust optimization, both in the structure of the problems they address and the size of the formulations. An algorithm for dealing with the problem with correlated uncertain parameters is also presented. Case studies in reverse production system infrastructure design are presented. The approach is generalizable to the robust design of network supply chain systems with reverse production systems as one of their subsystems.

<p>Assessments of chemical health risks are performed by scientific experts. Their intended use is as bases for decisions. This thesis tries to answer the questions of how uncertainty is, and should be, communicated in such risk assessments. The thesis consists of two articles and an introductory essay.</p><p>Article I focuses on the linguistic aspect of the communication of uncertainty in risk assessments. The aim of the article is to elucidate how risk assessors actually indicate uncertainty in risk assessment reports. Because of the prevalent uncertainty in risk assessment, deriving from several sources, uncertainty is communicated in verbal, rather than numerical terms. A typology of uncertainty indicators – phrases used to express uncertainty – is proposed and applied to the reviewed reports. It is found that the use of such phrases is not transparent, and the article concludes by a number of recommendations for improving the practice.</p><p>Article II mainly deals with the content of the communication. The overall question treated is what a characterization of uncertainty should include if a decision made on the basis of the risk assessment information is to be as well-founded as possible. A set of conditions is put forward to be fulfilled by a characterization of uncertainty if it is to be adequate from a decision-making point of view.</p><p>The greater part of the introductory essay is devoted to the concept of uncertainty which, at the conceptual level, does not appear to have been much discussed by philosophers</p>

Bayesian process control is a statistical process control (SPC) scheme that uses the posterior state probabilities as the control statistic. The key issue is to decide when to restore the process based on real-time observations. Such problems have been extensively studied in the framework of partially observable Markov decision processes (POMDP), with particular emphasis on the structure of optimal control policy. Almost all existing structural results on the optimal policies are limited to the two-state processes, where the class of control-limit policy is optimal. However, the two-state model is a gross simplification, as real production processes almost always involve multiple states. For example, a machine in the production system often has multiple failure modes differing in their effects; the deterioration process can often be divided into multiple stages with different degradation levels; the condition of a complex multi-unit system also requires a multi-state representation. We investigate the optimal control policies for multi-state processes with fixed sampling scheme, in which information about the process is represented by a belief vector within a high dimensional probability simplex. It is well known that obtaining structural results for such high-dimensional POMDP is challenging. Firstly, we prove that for an infinite-horizon process subject to multiple competing assignable causes, a so-called conditional control limit policy is optimal. The optimal policy divides the belief space into two individually connected regions, which have analytical bounds. Next, we address a finite-horizon process with at least one absorbing state and show that a structured optimal policy can be established by transforming the belief space into a polar coordinate system, where a so-called polar control limit policy is optimal. Our model is general enough to include many existing models in the literature as special cases. The structural results also lead to significantly efficient algorithms for computing the optimal policies. In addition, we characterize the condition for some out-of-control state to be more desirable than the in-control state. The existence of such counterintuitive situation indicates that multi-state process control is drastically different from the two-state case.

Bayesian process control is a statistical process control (SPC) scheme that uses the posterior state probabilities as the control statistic. The key issue is to decide when to restore the process based on real-time observations. Such problems have been extensively studied in the framework of partially observable Markov decision processes (POMDP), with particular emphasis on the structure of optimal control policy. Almost all existing structural results on the optimal policies are limited to the two-state processes, where the class of control-limit policy is optimal. However, the two-state model is a gross simplification, as real production processes almost always involve multiple states. For example, a machine in the production system often has multiple failure modes differing in their effects; the deterioration process can often be divided into multiple stages with different degradation levels; the condition of a complex multi-unit system also requires a multi-state representation. We investigate the optimal control policies for multi-state processes with fixed sampling scheme, in which information about the process is represented by a belief vector within a high dimensional probability simplex. It is well known that obtaining structural results for such high-dimensional POMDP is challenging. Firstly, we prove that for an infinite-horizon process subject to multiple competing assignable causes, a so-called conditional control limit policy is optimal. The optimal policy divides the belief space into two individually connected regions, which have analytical bounds. Next, we address a finite-horizon process with at least one absorbing state and show that a structured optimal policy can be established by transforming the belief space into a polar coordinate system, where a so-called polar control limit policy is optimal. Our model is general enough to include many existing models in the literature as special cases. The structural results also lead to significantly efficient algorithms for computing the optimal policies. In addition, we characterize the condition for some out-of-control state to be more desirable than the in-control state. The existence of such counterintuitive situation indicates that multi-state process control is drastically different from the two-state case.

Cette thèse contient trois essais expérimentaux étudiant les causes possibles de l’inefficience des marchés des actifs. L’efficacité des marchés financiers est cruciale pour une bonne performance de l’économie dans son ensemble. La recherche en finance comportementale a montré que les investisseurs ne se comportent pas toujours de manière parfaitement rationnelle. Il est donc important de bien comprendre comment les individus créent leurs croyances concernant les décisions financières, ce qui les influence, comment elles affectent les marchés financiers, et donc l’efficacité des marchés.Les croyances individuelles relatives à une décision financière sont influencées par la façon dont les actifs sont déterminés. Le premier essai étudie l’impact du degré d’asymétrie des actifs échangés sur : premièrement, le développement du marché global, deuxièmement, la façon dont les individus perçoivent les actifs risqués en fonction de leurs préférences de risque, et troisièmement, la stabilité de la perception du risque de ces actifs dans le temps. Nos résultats suggèrent que l’asymétrie des actifs n’influence que marginalement le développement du marché, mais a un effet direct sur la perception du risque. Les décisions des agents qui interagissent sur les marchés financiers sont influencées par leurs préférences, leurs traits de personnalité et leurs biais comportementaux. Nous supposons que le profil personnel influe aussi bien sur le comportement individuel sur le marché, tels que l’activité d’échange, l’accumulation de stock et la performance, que sur le développement du marché global, comme la dynamique du prix ou le nombre d’actifs échangés. C’est l’objectif du deuxième essai. Nous constatons que les traits de personnalité sont les meilleurs prédicateurs de comportement du marché, à la fois individuel et global. Le troisième essai examine l’impact des incitations concurrentielles sur l’augmentation des anomalies de marché. Dans ce cas, allonger l’échelle de temps sur laquelle les comparaisons des performances sont basées, contribue-t-il à améliorer l’efficience des marchés financiers ? Nous constatons que le bonus à l’échelle de temps étendue aidera à réduire les anomalies du marché et à améliorer l’efficacité du marché financier. / This thesis contains three essays that focus on asset market inefficiency using the experimental method. Financial market efficiency is crucial for good performance of the economy as a whole. Research in behavioral finance has shown that investors do not always behave fully rationally and systematically violate the assumptions of the traditional framework. It is therefore important to fully understand how individuals create their expectations regarding financial decisions, what influences them, how they affect the global market, and therefore financial market efficiency.Individual expectations about a financial decision are influenced by the manner assets are determined. The first essay investigates the impact of skewness of traded assets on first, aggregate market development, second, the way individuals perceive risky assets according to their risk preferences, and third, the stability of the assets’ risk perception in time. Our results suggest that assets’ skewness influences only marginally the asset market development, but directly effects the individual risk perception.Agents interacting in financial markets are not fully rational. Their decisions are influenced by their preferences, personality traits and the degree they are prone to behavioral biases. We suppose that the personal profile influences individual market behavior, such as trading activity, stock accumulation and performance, and also the aggregate market development, such as price dynamic or turnover of traded assets. This is the objective of the second essay. We find that the personality traits are the best predictors of both individual and aggregate market behavior.The third essay examines whether competitive incentives do contribute to the increase of mispricing in financial markets. If they do, does the extended time horizon of performance comparison help to improve the control against excessive risk-taking and therefore improve financial market efficiency. We find that the bonuses with extended time horizon help to diminish mispricing and improve the financial market efficiency.