Risk-based regulation has grown rapidly as a component of Government decision making, and as such, the need for an established evidence-based framework for decisions about risk has become the new mantra. However, the process of brokering scientific evidence is poorly understood and there is a need to improve the transparency of this brokering process and decisions made. This thesis attempts to achieve this by using agent-based simulation to model the influence that power structures and participating personalities has on the brokering of evidence and thereby the confidence-building exercise that characterises risk-based regulation. As a prerequisite to the adoption of agent-based techniques for simulating decisions under uncertainty, this thesis provides a critical review of the influence power structure and personality have on the brokering of scientific evidence that informs risk decisions. Three case studies, each representing a different perspective on risk-based regulation are presented: nuclear waste disposal, the disposal of avian-influenza infected animal carcases and the reduction of dietary salt intake. Semi-structured interviews were conducted with an expert from each case study, and the logical sequence in which decisions were made was mapped out and used to inform the development of an agent-based simulation model. The developed agent-based model was designed to capture the character of the brokering process by transparently setting out how evidence is transmitted from the provider of evidence to the final decision maker. It comprises of two agents, a recipient and provider of evidence, and draws upon a historic knowledge base to permit the user to vary components of the interacting agents and of the decision-making procedure, demonstrating the influence that power structure and personality has on agent receptivity and the confidence attached to a number of different lines of evidence. This is a novel step forward because it goes beyond the scope of current risk management frameworks, for example, permitting the user to explore the influence that participants have in weighing and strengthening different lines of evidence and the impact this has on the final decision outcome.

Customer retention is an important issue for any business, especially in mature markets such as the UK mobile market where new customers can only be acquired from competitors. Different methods and techniques have been used to investigate customer retention including statistical methods and data mining. However, due to the increasing complexity of the mobile market, the effectiveness of these techniques is questionable. This study proposes Agent-Based Modelling and Simulation (ABMS) as a novel approach to investigate customer retention. ABMS is an emerging means of simulating behaviour and examining behavioural consequences. In outline, agents represent customers and agent relationships represent processes of agent interaction. This study follows the design science paradigm to build and evaluate a generic, reusable, agent-based (CubSim) model to examine the factors affecting customer retention based on data extracted from a UK mobile operator. Based on these data, two data mining models are built to gain a better understanding of the problem domain and to identify the main limitations of data mining. This is followed by two interrelated development cycles: (1) Build the CubSim model, starting with modelling customer interaction with the market, including interaction with the service provider and other competing operators in the market; and (2) Extend the CubSim model by incorporating interaction among customers. The key contribution of this study lies in using ABMS to identify and model the key factors that affect customer retention simultaneously and jointly. In this manner, the CubSim model is better suited to account for the dynamics of customer churn behaviour in the UK mobile market than all other existing models. Another important contribution of this study is that it provides an empirical, actionable insight on customer retention. In particular, and most interestingly, the experimental results show that applying a mixed customer retention strategy targeting both high value customers and customers with a large personal network outperforms the traditional customer retention strategies, which focuses only on the customer‘s value.

Agent-based models (ABM) are an increasingly important research tool for describing and predicting interactions among humans and their environment. A key challenge for such models is the ability to faithfully represent human decision making with respect to observed behaviour. This thesis aims to address this challenge by developing a methodology for empirical measurement and simulation of decision making in humanenvironment systems. The methodology employs the Beliefs-Desires-Intentions (BDI) model of human reasoning to directly translate empirically measured decision data into artificial agents, based on sound theoretical principles. A common simulated decision environment is used for both eliciting human decision making behaviour, and validating artificial agents. Using this approach facilitates the collection of decision making narratives by way of participatory simulation, and promotes a fair comparison of real and modelled decision making. The methodology is applied in two case studies: One to carry out a trial involving human subjects solving an abstract land-use problem, and another to examine the feasibility of up-scaling the methodology to a real agricultural scenario—dairy farming. Results from the experiments indicate that the BDI-based methodology achieved reasonably direct encoding of decision making behaviour from elicited human narratives. The main limitations found with the technique are: (1) the significant use of subjects’ time required to elicit their decision making behaviour; (2) the significant programming effort required; and (3) the challenge of aggregating behaviour from multiple subjects into a generalised decision making model. In spite of its limitations, BDI has shown its strengths as a tool for empirical analysis and simulation of decision making in research of human-environment systems.

The future electricity grid is likely to be increasingly complex and uncertain due to the introduction of new technologies in the grid, the increased use of control and communication infrastructure, and the uncertain political climate. In recent years, the transactive energy market framework has emerged as the key framework for future electricity market design in the electricity grid. However, most of the work done in this area has focused on developing retail level transactive energy markets. There seems to be an underlying assumption that wholesale electricity markets are ready to support any retail market design. In this dissertation, we focus on designing wholesale electricity markets that can better support transactive retail market. On the highest level, this dissertation contributes towards developing tools and models for future electricity market designs. A particular focus is placed on the relationship between wholesale markets and investment planning. Part I of this dissertation uses relatively simple models and case studies to evaluate key impediments to flexible transmission operation. In doing so, we identify several potential areas of concern in wholesale market designs: 1. There is a lack of consideration of demand flexibility both in the long-run and in the short-run 2. There is a disconnect between operational practices and investment planning 3. There is a need to rethink forward markets to better manage resource adequacy under long-term uncertainties 4. There is a need for more robust modeling tools for wholesale market design In Part II and Part III of this dissertation, we make use of mathematical decomposition and agent-based simulations to tackle these concerns. Part II of this dissertation uses Benders Decomposition and Lagrangian Decomposition to spatially and temporally decompose a power system and operation problem with active participation of flexible loads. In doing so, we are able to not only improve the computational efficiency of the problem, but also gain various insights on market structure and pricing. In particular, the decomposition suggests the need for a coordinated investment market and forward energy market to bridge the disconnect between operational practices and investment planning. Part III of this dissertation combines agent-based modeling with state-machine based modeling to test various spot, forward, and investment market designs, including the coordinated investment market and forward energy market proposed in Part II of this dissertation. In addition, we test a forward energy market design where 75% of load is required to be purchased in a 2-year-ahead forward market and various transmission cost recovery strategies. We demonstrate how the different market designs result in different investment decisions, winners, and losers. The market insights lead to further policy recommendations and open questions. Overall, this dissertation takes initial steps towards demonstrating how mathematical decomposition and agent-based simulations can be used as part of a larger market design toolbox to gain insights into different market designs and rules for the future electricity grid. In addition, this dissertation identifies market design ideas for further studies, particularly in the design of forward markets and investment cost recovery mechanisms.

With increasing computational power and more elaborate software comes greater opportunities to complement traditional research methods with alternative methods. In this paper we argue for why the area of two-sided markets could benefit from this alternative approach and attempt to implement a theoretical model in an agent-based framework. By first replicating the theoretical findings in this framework we expand the model in increments in different directions through introducing different set of heterogeneity and behavioral limitations on our actors to see how the theoretical model develops. Only changing the model in increments found the analytical outcome to be robust for many of our changes, in this regard we have not managed to successfully take advantage of the full potential of the agent-based framework.

Widespread unexplained variations in clinical practices and patient outcomes, together with rapidly growing availability of data, suggest major opportunities for improving the quality of medical care. One way that healthcare practitioners try to do that is by participating in organized healthcare quality improvement collaboratives (QICs). In QICs, teams of practitioners from different hospitals exchange information on clinical practices, with the aim of improving health outcomes at their own institutions. However, what works in one hospital may not work in others with different local contexts, due to non-linear interactions among various demographics, treatments, and practices. I.e., the clinical landscape is a complex socio-technical system that is difficult to search. In this dissertation we develop methods for analysis and modeling of complex systems, and apply them to the problem of healthcare improvement. Searching clinical landscapes is a multi-objective dynamic problem, as hospitals simultaneously optimize for multiple patient outcomes. We first discuss a general method we developed for finding which changes in features may be associated with various changes in outcomes at different points in time with different delays in affect. This method correctly inferred interactions on synthetic data, however the complexity and incompleteness of the real hospital dataset available to us limited the usefulness of this approach. We then discuss an agent-based model (ABM) of QICs to show that teams comprising individuals from similar institutions outperform those from more diverse institutions, under nearly all conditions, and that this advantage increases with the complexity of the landscape and the level of noise in assessing performance. We present data from a network of real hospitals that provides encouraging evidence of a high degree of similarity in clinical practices among hospitals working together in QIC teams. Based on model outcomes, we propose a secure virtual collaboration system that would allow hospitals to efficiently identify potentially better practices in use at other institutions similar to theirs, without any institutions having to sacrifice the privacy of their own data. To model the search for quality improvement in clinical fitness landscapes, we need benchmark landscapes with tunable feature interactions. NK landscapes have been the classic benchmarks for modeling landscapes with epistatic interactions, but the ruggedness is only tunable in discrete jumps. Walsh polynomials are more finely tunable than NK landscapes, but are only defined on binary alphabets and, in general, have unknown global maximum and minimum. We define a different subset of interaction models that we dub as NM landscapes. NM landscapes are shown to have smoothly tunable ruggedness and difficulty and known location and value of global maxima. With additional constraints, we can also determine the location and value of the global minima. The proposed NM landscapes can be used with alphabets of any arity, from binary to real-valued, without changing the complexity of the landscape. NM landscapes are thus useful models for simulating clinical landscapes with binary or real decision variables and varying number of interactions. NM landscapes permit proper normalization of fitnesses so that search results can be fairly averaged over different random landscapes with the same parameters, and fairly compared between landscapes with different parameters. In future work we plan to use NM landscapes as benchmarks for testing various algorithms that can discover epistatic interactions in real world datasets.

Multiple research problems related to the lung involve a need to take into account the spatiotemporal dynamics of the underlying component cells. Two such problems involve better understanding the nature of the allergic inflammatory response to explore what might cause chronic inflammatory diseases such as asthma, and determining the rules underlying stem cells used to engraft decellularized lung scaffolds in the hopes of growing new lungs for transplantation. For both problems, we model the systems computationally using agent-based modeling, a tool that enables us to capture these spatiotemporal dynamics by modeling any biological system as a collection of agents (cells) interacting with each other and within their environment. This allows to test the most important pieces of biological systems together rather than in isolation, and thus rapidly derive biological insights from resulting complex behavior that could not have been predicted beforehand, which we can then use to guide wet lab experimentation. For the allergic response, we hypothesized that stimulation of the allergic response with antigen results in a response with formal similarity to a muscle twitch or an action potential, with an inflammatory phase followed by a resolution phase that returns the system to baseline. We prepared an agent-based model (ABM) of the allergic inflammatory response and determined that antigen stimulation indeed results in a twitch-like response. To determine what might cause chronic inflammatory diseases where the twitch presumably cannot resolve back to baseline, we then tested multiple potential defects to the model. We observed that while most of these potential changes lessen the magnitude of the response but do not affect its overall behavior, extending the lifespan of activated pro-inflammatory cells such as neutrophils and eosinophil results in a prolonged inflammatory response that does not resolve to baseline. Finally, we performed a series of experiments involving continual antigen stimulation in mice, determining that there is evidence in the cytokine, cellular and physiologic (mechanical) response consistent with our hypothesis of a finite twitch and an associated refractory period. For stem cells, we made a 3-D ABM of a decellularized scaffold section seeded with a generic stem cell type. We then programmed in different sets of rules that could conceivably underlie the cell's behavior, and observed the change in engraftment patterns in the scaffold over selected timepoints. We compared the change in those patterns against the change in experimental scaffold images seeded with C10 epithelial cells and mesenchymal stem cells, two cell types whose behaviors are not well understood, in order to determine which rulesets more closely match each cell type. Our model indicates that C10s are more likely to survive on regions of higher substrate while MSCs are more likely to proliferate on regions of higher substrate.

Healthcare systems are traditionally characterised by complexity and heterogeneity. With the continuous increase in size and shrinkage of available resources, the healthcare sector faces the challenge of delivering high quality services with fewer resources. Healthcare organisations cannot be seen in isolation since the services of one such affects the performance of other healthcare organisations. Efficient management and forward planning, not only locally but rather across the whole system, could support healthcare sector to overcome the challenges. An example of closely interwoven organisations within the healthcare sector is the emergency medical services (EMS). EMS operate in a region and usually consist of one ambulance service and the available accident and emergency (A&E) departments within the coverage area. EMS provide, mainly, pre-hospital treatment and transport to the appropriate A&E units. The life-critical nature of EMS demands continuous systems improvement practices. Modelling and Simulation (M&S) has been used to analyse either the ambulance services or the A&E departments. However, the size and complexity of EMS systems constitute the conventional M&S techniques inadequate to model the system as a whole. This research adopts the approach of distributed simulation to model all the EMS components as individual and composable simulations that are able to run as standalone simulation, as well as federates in a distributed simulation (DS) model. Moreover, the hybrid approach connects agent-based simulation (ABS) and discrete event simulation (DES) models in order to accommodate the heterogeneity of the EMS components. The proposed FIELDS Framework for Integrated EMS Large-scale Distributed Simulation supports the re-use of existing, heterogeneous models that can be linked with the High Level Architecture (HLA) protocol for distributed simulation in order to compose large-scale simulation models. Based on FIELDS, a prototype ABS-DES distributed simulation EMS model was developed based on the London EMS. Experiments were conducted with the model and the system was tested in terms of performance and scalability measures to assess the feasibility of the proposed approach. The yielded results indicate that it is feasible to develop hybrid DS models of EMS that enables holistic analysis of the system and support model re-use. The main contributions of this thesis is a distributed simulation methodology that derived along the process of conducting this project, the FIELDS framework for hybrid EMS distributed simulation studies that support re-use of existing simulation models, and a prototype distributed simulation model that can be potentially used as a tool for EMS analysis and improvement.

This study asks the question: If we take a small neighborhood and introduce a barrier, how will the neighborhood change? Will it be better protected and flourish, or will it decay and die or perhaps will there be no change at all? What determines the outcome? This work tries to answer these questions by creating an Agent-Based Model (ABM) to test different scenarios and observe the results. Urban environments, both natural and built, are complex systems, containing a multitude of people, landscapes and buildings. Simple changes in street-lighting and sidewalks, the addition of trees and green scapes or the enforcement of “broken-window” policies impact local neighborhoods [1]. Measuring behavior changes on a neighborhood level are difficult to quantify, but by using ABM methods we can build our neighborhood, populate it with a variety of actors and watch their interaction with each other, and with introduced stimuli. Our simulation introduces a barrier (highway) with varying permeability into a mixed use neighborhood loosely based on Richmond’s Jackson Ward. Several metrics (such as property value, crime rates, etc.) were used to determine if the neighborhood was under duress, or thriving. In real-world terms we built a roadway though the neighborhood and observed the “severance effect” as our actors’ adapted to reduced mobility and remained within their accessible range.

From the agent-based, correlated random walk model presented, we observe the effects of varying the parameter values of maximum insect turning area, 𝛿max, density of trees, ω, maximum pollen carryover, 𝜅max, and probability of fertilization, P𝜅, on the distribution of pollen within a population of Cornus florida (flowering dogwood). We see that varying 𝛿max and 𝜅max changes the dispersal distance of pollen, which greatly affects many measures of connectivity. The clustering coefficient of fathers is maximized when 𝛿max is between 60° and 90°. Varying ω does not have a major effect on the clustering coefficient of fathers, but it does have a greater effect on other measures of genetic diversity. Lastly, we compare our simulations with randomly-placed trees with that of actual tree placement of C. florida at the VCU Rice Center, concluding that in order to truly understand how pollen is distributed within a specific ecosystem, specificity in describing tree locations is necessary.