This dissertation concerns the self-organization behaviors in different types of multi-agent systems, and possible ways to apply interventions on top ofthat to achieve certain goals. A bounded confidence opinion dynamics modelis considered for the first two papers. Theoretical analysis of the model isperformed and modifications of the model are given so that it will have better properties in some aspect. Leader-follower based models are studied in the third to fifth papers where various optimal control problems are considered. Different methods such as Pontryagin minimum principle and dynamic programming are used to solve those optimal control problem. For complex problems, one may only get approximate solutions or suboptimal solutions.In Paper A and Paper B, we consider the continuous-time Hegselmann-Krause (H-K) model and its variations and target the problem of reaching consensus. A sufficient condition on the initial opinion distribution is givento guarantee consensus for the original continuous-time H-K model. A modified model is provided and proven to be able to lead a larger range of initial opinions to synchronization. An H-K model with an exo-system is also studied where sufficient conditions on the exo-system are given for the purpose of consensus.In Paper C and Paper D, optimal control problems with leader-followerbased multi-agent systems are discussed. Analytic solutions are derived if the dynamics is linear by applying Pontryagin minimum principle. For generalnon-linear leader-follower interactions, we provide a method that use sstatistic moments of the follower crowd to approximate the optimal control.The dynamic programming approach is used and certain approximation ofthe Hamilton-Jacobi-Bellman equations is needed. The computational burdenis so heavy that model predictive control method is required in practical applications.In Paper E, we apply a similar method to the approach used in PaperD to target a pollutant elimination problem. It implies that we can use themethod to attack optimal control problem with partial differential equation constraints by discretization in space. The dimension of the discretization is not related to the computational complexity since only the statistic moments are needed. / <p>QC 20161201</p>

Les approches habituelles pour la protection de la privacité s'attachent à définir un niveau de sensibilité pour chaque information. Cette information est soit publique, soit privée et sa circulation est restreinte à un groupe d'agents prédéfini. Dans cette thèse, nous nous appuyons sur la théorie de l'intégrité contextuelle, qui propose de redéfinir la notion de violation de privacité. Selon cette théorie, toute transmission peut déclencher une violation de privacité suivant le contexte dans lequel elle a lieu. Cette thèse utilise la théorie de l'intégrité contextuelle afin de proposer un modèle de protection de la privacité pour les systèmes multi-agents décentralisés: le modèle PrivaCIAS. Afin de contrôler les agents dans le système, le modèle PrivaCIAS fournit un ensemble de normes qui permet la mise en place d'un contrôle social basé sur la confiance. Le modèle donne le contrôle aux agents pour constater les violations (selon l'intégrité contextuelle), puis punir les contrevenants en les excluant du système sans avoir besoin de recourir à une autorité centrale. Ce modèle vise les réseaux sociaux décentralisés comme champ d'application. / Contextual Integrity has been proposed to define privacy in an unusual way. Most approaches take into account a sensitivity level or a ``privacy circle'': the information is said to be either private or public and to be constrained to a given group of agents, \textit{e.g.} ``my friends'', when private. In the opposite, Contextual Integrity states that any information transmitted can make this transmission a privacy violation depending on its context. In this thesis, we use this theory to develop a novel model that one can use in an open and decentralized virtual community to socially enforce privacy. This thesis defines the PrivaCIAS model, in which privacy constraints are formally described to be used to detect privacy violations according to the Contextual Integrity theory. The PrivaCIAS model provides norms to agents in order to make them implement social control. The model does not require a central authority, it gives control to the agents for detecting privacy violations (through Contextual Integrity) and excluding violating agents from the system through social exclusion. This model targets decentralized social networks as a main application domain.

Approved for public release, distribution is unlimited / The delivery of supply in combat operations is very important and often results in success or failure of a mission. This activity, as well as other transportation problems, has traditionally been modeled using global optimization techniques, such as linear programming. However, the goal of this thesis is to examine the feasibility of an agent-based solution to study the movement of material through a road network. The requirement is to build an agent-based system that finds the optimal route through a given road network and is capable of adapting to disruptions introduced to the network and then find alternative routes through the network. The agents act from a local perspective, and can represent more realistically the decisions being made throughout the delivery process. This thesis implements an analysis environment for road networks and develops an agent-based model to build truck-driver agents that are capable of delivering supplies through a changing road network. / Captain, German Army

This thesis introduces a financial simulation model covering two related financial domains: Mobile Payments and Retail Stores systems.   The problem we address in these domains is different types of fraud. We limit ourselves to isolated cases of relatively straightforward fraud. However, in this thesis the ultimate aim is to introduce our approach towards the use of computer simulation for fraud detection and its applications in financial domains. Fraud is an important problem that impact the whole economy. Currently, there is a lack of public research into the detection of fraud. One important reason is the lack of transaction data which is often sensitive. To address this problem we present a mobile money Payment Simulator (PaySim) and Retail Store Simulator (RetSim), which allow us to generate synthetic transactional data that contains both: normal customer behaviour and fraudulent behaviour.    These simulations are Multi Agent-Based Simulations (MABS) and were calibrated using real data from financial transactions. We developed agents that represent the clients and merchants in PaySim and customers and salesmen in RetSim. The normal behaviour was based on behaviour observed in data from the field, and is codified in the agents as rules of transactions and interaction between clients and merchants, or customers and salesmen. Some of these agents were intentionally designed to act fraudulently, based on observed patterns of real fraud. We introduced known signatures of fraud in our model and simulations to test and evaluate our fraud detection methods. The resulting behaviour of the agents generate a synthetic log of all transactions as a result of the simulation. This synthetic data can be used to further advance fraud detection research, without leaking sensitive information about the underlying data or breaking any non-disclose agreements.   Using statistics and social network analysis (SNA) on real data we calibrated the relations between our agents and generate realistic synthetic data sets that were verified against the domain and validated statistically against the original source.   We then used the simulation tools to model common fraud scenarios to ascertain exactly how effective are fraud techniques such as the simplest form of statistical threshold detection, which is perhaps the most common in use. The preliminary results show that threshold detection is effective enough at keeping fraud losses at a set level. This means that there seems to be little economic room for improved fraud detection techniques.   We also implemented other applications for the simulator tools such as the set up of a triage model and the measure of cost of fraud. This showed to be an important help for managers that aim to prioritise the fraud detection and want to know how much they should invest in fraud to keep the loses below a desired limit according to different experimented and expected scenarios of fraud.

[pt] A abordagem de sistemas auto-organizáveis vem crescendo em relevância e uso no âmbito de sistemas complexos, pois permite o desenvolvimento de sistemas descentralizados que exibem uma dinamicidade e adaptabilidade em consequência de perturbações no ambiente, anteriormente desconhecidas. As principais dificuldades na construção de sistemas auto-organizáveis residem na elaboração de mecanismos de interação e coordenação entre os agentes do ambiente e na falta do reúso de soluções consagradas. Nesse contexto, esta dissertação propõe um framework como solução reutilizável para a construção de sistemas auto-organizáveis descentralizados, baseando-se nos principais padrões arquiteturais encontrados na literatura, proporcionando também um meio de extensibilidade para elaboração de novos mecanismos de interação e coordenação. A partir do framework, instâncias de diversos domínios podem ser criadas, como exemplo, uma solução auto-organizável e descentralizada para os veículos guiados automaticamente, como será apresentada nesta dissertação. / [en] The approach of self-organizing systems has increased in relevance and use within complex domains, for it allows the development of decentralized systems that exhibit a dynamic and adaptable behavior in facing the challenge of handling disturbances in the environment, which were previously unknown. The main difficulties in building self-organizing systems lie in the development of mechanisms of interaction and coordination between the agents of the environment and the lack of reuse of solutions already adopted. In this context, this dissertation proposes a framework as a reusable solution for building decentralized self-organizing systems, based on major architectural patterns found in the literature, and also provide a means of extensibility to develop new mechanisms of interaction and coordination. From the framework, instances of various fields can be created, for example, a self-organizing and decentralized solution to the automated guided vehicles problem, as will be presented in this dissertation.

[pt] Simulações de Ambientes Virtuais (VES) são um tipo especial de simulação, normalmente utilizado para implementar jogos ou jogos sérios com representação em espaço virtual e utilizando a técnicas de avanço do tempo de simulação para o próximo evento ou dirigida por tempo. Um exemplo de uso é a aplicação em simulações de apoio a Jogos de Guerra. O uso de Simulações Baseadas em Sistemas Multiagentes (MABS) para implementar VES é adequado devido a possibilidade de modelar e implementar o sistema com ênfase nos atores e seus comportamentos dinâmicos. Contudo, quando os agentes da simulação passam a controlar o avanço do tempo de simulação de forma individualizada, surge uma situação semelhante á simulação paralela, o que implica na necessidade de tratar questões como disputa por recursos computacionais pelos agentes, atrasos em tempo de execução e a perda de consistência da própria simulação. A situação torna-se mais complicada em sistemas desenvolvidos com Java, devido a particularidades desta tecnologia. Este trabalho apresenta uma abordagem para controlar estes atrasos em tempo de execução e assim apoiar o desenvolvimento de VES utilizando MABS com Java. / [en] Virtual Environment Simulations (VES) are a special type of simulation, often used to implement games and serious games with virtual space representation and using both the next-event or stepped–time simulation time advance approach. An example of serious games is the simulation used to support War Games. Multiagent Based Simulation (MABS) are suitable to implement these simulations because of their ability to handle with complexity and individual actors modeling. However, when agents are responsible for advance their own simulation time, a situation similar to a parallel simulation arises. This implies in treat issues such as delays in performing scheduled actions (i.e tardiness) and its consequences in the virtual space representations. This situation is worst in Java based MABS, because of Java technology particularities. This work presents an approach to tame this tardiness and help the development of these cited VES using agent oriented paradigm.

Le monde des entreprises est profondément soumis à un ensemble de contraintes toujours plus exigeantes provenant d’une part des clients, exigeant des produits plus personnalisables, de qualité supérieure et à faible coût, et d’autre part des aléas internes auxentreprises, comprenant les pannes machines, les défaillances humaines, la fluctuation de la demande, les fréquentes variations de production. Cette thèse propose une architecture de contrôle de systèmes de production, basée sur les principes holoniques développées dans l’architecture ADACOR (ADAptive holonic COntrol aRchitecture), et l’étendant en s’inspirant des théories de l’évolution et en utilisant des mécanismes d’auto-organisation. L’utilisation des théories de l’évolution enrichit l’architecture de contrôle en permettant l’évolution de deux manières distinctes, en réponse au type et au degré de la perturbation apparue. Le premier mode d’adaptation, appelé auto-organisation comportementale, permet à chaque entité qui compose le système d’adapter dynamiquement leur comportement interne, gérant de cette façon de petites perturbations. Le second mode, nommé auto-organisation structurelle, traite de plus grandes perturbations, en permettant aux entités du système de ré-organiser leurs relations, et par conséquent modifier structurellement le système. L’architecture holonique auto-organisée de contrôle de systèmes de production proposée dans cette thèse a été validée sur une cellule de production flexible AIP-PRIMECA. Les résultats ont montré une amélioration des indicateurs clés de performance par rapport aux architectures de contrôle hiérarchiques et hétérarchiques. / The manufacturing world is being deeply challenged with a set of ever demanding constraints where from one side, the costumers are requiring products to be more customizable, with higher quality at lower prices, and on other side, companies have to deal on a daily basis with internal disturbances that range from machine breakdown to worker absence and from demand fluctuation to frequent production changes. This dissertation proposes a manufacturing control architecture, following the holonic principles developed in the ADAptive holonic COntrol aRchitecture (ADACOR) and extending it taking inspiration in evolutionary theories and making use of self- organization mechanisms. The use of evolutionary theories enrich the proposed control architecture by allowing evolution in two distinct ways, responding accordingly to the type and degree of the disturbance that appears. The first component, named behavioural self- organization, allows each system’s entity to dynamically adapt its internal behaviour, addressing small disturbances. The second component, named structural self-organization, addresses bigger disturbances by allowing the system entities to re-arrange their rela- tionships, and consequently changing the system in a structural manner. The proposed self-organized holonic manufacturing control architecture was validated at a AIP-PRIMECA flexible manufacturing cell. The achieved experimental results have also shown an improvement of the key performance indicators over the hierarchical and heterarchical control architecture.

[pt] Sistemas multi-agentes abertos são freqüentemente caracterizados como sistemas distribuídos onde agentes, que podem não ser conhecidos a priori, podem entrar ou sair do sistema a qualquer momento. Uma abordagem de governança estabelece regras de interação que precisam ser seguidas pelos agentes de um determinado sistema. O uso destas regras permite uma maior previsibilidade e controle sobre o comportamento observável do sistema. Nesta tese, apresenta-se uma abordagem de governança de sistemas multi-agentes abertos como adequada para lidar não apenas com o monitoramento e controle do comportamento dos agentes, mas também com aspectos de fidedignidade. Um software é dito fidedigno quando se pode confiar no mesmo através de verificações formais ou informais assumindo riscos de danos compatíveis com o serviço prestado pelo software. A incorporação de aspectos de fidedignidade em uma abordagem de governança tem como principal benefício a geração de uma tecnologia integrada que possui as vantagens de uma abordagem de governança e ao mesmo tempo lida com especificações de instrumentos de fidedignidade tais como prevenção e tolerância a faltas. A abordagem proposta é ilustrada através de um estudo de caso no contexto de controle de tráfego aéreo. / [en] Open multi-agent systems are frequently characterized by having little or no control over the behavior of the agents. The internal implementation and architecture of agents usually are inaccessible, and different teams may have developed them but with no coordination between them. Furthermore, agents may enter or leave the system at their will. A governance approach defines the interaction rules that must be obeyed by the agents. These rules allow for a greater control and predictability of the observable system behavior. In this thesis, we propose a governance approach to deal not only with the monitoring and control of agents` behavior but also to deal with dependability concerns. The original definition of dependability is the ability to deliver service that can justifiably be trusted. A governance approach that also deals with dependability has as main benefit the reuse of the monitoring and enforcement present in the governance infrastructure for dependability. We present a case study in the context of an air traffic control system to illustrate our approach.

[pt] Mecanismos de reputação são utilizados para aumentar a confiança e o desempenho de sociedades virtuais. Diferentes modelos de reputação descentralizados foram propostos baseados em interações entre agentes. Cada agente de software é capaz de avaliar e armazenar as reputações dos agentes com quem eles interagiram e também podem testemunhar a outros agentes sobre tais reputações. As desvantagens principais de tais abordagens quando aplicadas em sistemas multi- agentes abertos e de larga escala são a dificuldade de estabelecer interações repetidas entre os agentes, a inviabilidade, em algumas situações, do processo de busca por testemunhas, o fato de a reputação ser influenciada pelo ponto de vista de um outro agente e o fato de que os agentes podem não estar dispostos a testemunhar colaborando com agentes concorrentes. Esta dissertação propõe um modelo de reputação híbrido, reunindo características de abordagens centralizada e descentralizada para superar tais problemas. As reputações são providas pelos próprios agentes de software mas também por subsistemas centralizados que podem ser facilmente acessados por qualquer agente e podem fornecer reputações seguras baseadas em testemunhos sobre comportamentos indesejados dos agentes, caracterizados pelas violações de normas do sistema. Tais subsistemas centralizados são instâncias do framework proposto, que flexibiliza as fórmulas de cálculo da reputação, o tempo de atualização da mesma e permite criar novos tipos de reputações relacionados a contextos diferentes. / [en] Reputation mechanisms are being used to increase the reliability and performance of virtual societies. Different decentralized reputation models have been proposed based on interactions among agents. Each system agent evaluates and stores the reputations of the agents with whom they have interacted and can also testify to other agents about such reputations. The main disadvantages of such approaches when applied to open large-scale multi-agent systems are the difficulty of establishing strong links between the agents, the sometimes infeasible witness search process, the fact of the reputation is being influenced by the point of view of another agent and the fact that the agents cannot be willing to testify and collaborate with possible competitive agents. In this work we propose a hybrid reputation system with centralized and decentralized characteristics to overcome such problems. The reputations are provided by the system agents themselves but also by centralized subsystems that can be easily reached by any agent and can supply reliable reputations of any agent based on testimonies about undesired agent's behavior, characterized by the violation of system norms. Such centralized subsystems are instances of the proposed framework.

[pt] Nos dias atuais, muito mais do que no passado, é sabido que o sucesso de projetos de software depende criticamente de requisitos. Um exemplo de GORE, que significa Engenharia de Requisitos Orientada a Metas, é o Framework i* (iestrela). Este afirma que os requisitos devem representar a intencionalidade do grande número de atores sociais, os quais são pessoas ou sistemas. Dentro desse contexto, existe ainda um vazio de métodos para cobrir o processo da elicitação de metas. Vários métodos para SMA (Sistemas Multi-Agentes) mencionam a elicitação das metas, mas eles não fornecem detalhes de como essa atividade é feita, eles tratam principalmente da modelagem de metas sem antes cuidar dos detalhes da elicitação. Em adição, modelos intencionais como exemplo o Framework i*, podem ser complexos e até mesmo incompreensíveis. Esta tese propõe um método chamado ERi*c – “Engenharia de Requisitos Intencional”, o qual primeiro provê um processo de elicitação que aplica uma abordagem “bottom-up” e simples, com perguntas que podem identificar metas concretas e metas flexíveis. Além disso, o método Eri*c propõe uma solução para reduzir o problema da “escalabilidade” de modelos i* que inclui heurísticas para a especificação desses modelos. O método contém também uma estratégia para fazer diagnósticos de modelos i* de maneira a cobrir o processo de análise de requisitos. / [en] Nowadays, much more than in the past, it is known that the success of software projects depends critically on the requirements. Goal Oriented Requirements Engineering - GORE, for example i Framework, says that requirements must represent the intentionality of a large number of social actors, which can be people or systems. Several Multi-Agent Systems (MAS) methods mention goals elicitation but they do not provide details of how this is performed, they mainly focus on goals modeling. In this context, there is still a lack of methods to cover the goal elicitation process. Only after eliciting goals, requirements engineers will be able to deal properly with goal models. Typically, this is a difficult task to carry on since requirements engineers are not familiarized with the domain from the early stages of software development. And, intentionality models, for example i Framework, can be complex and incompreensible. This thesis proposes a method called ERi c - Engenharia de Requisitos Intencional which provides an inquire process that can identify goals and softgoals in a bottom-up and simple elicitation approach together with one solution to reduce the problem of scalability of i models. The method ERi c also includes heuristics for modeling specification and a diagnoses aproach in order to analyze i models.