Reinforcement learning (RL) has received attention in recent years from agent-based researchers because it can be applied to problems where autonomous agents learn to select proper actions for achieving their goals based on interactions with their environment. Each time an agent performs an action, the environment¡Šs response, as indicated by its new state, is used by the agent to reward or penalize its action. The agent¡Šs goal is to maximize the total amount of reward it receives over the long run. Although there have been several successful examples demonstrating the usefulness of RL, its application to manufacturing systems has not been fully explored. The objective of this research is to develop a set of guidelines for applying the Q-learning algorithm to enable an individual agent to develop a decision making policy for use in agent-based production scheduling applications such as dispatching rule selection and job routing. For the dispatching rule selection problem, a single machine agent employs the Q-learning algorithm to develop a decision-making policy on selecting the appropriate dispatching rule from among three given dispatching rules. In the job routing problem, a simulated job shop system is used for examining the implementation of the Q-learning algorithm for use by job agents when making routing decisions in such an environment. Two factorial experiment designs for studying the settings used to apply Q-learning to the single machine dispatching rule selection problem and the job routing problem are carried out. This study not only investigates the main effects of this Q-learning application but also provides recommendations for factor settings and useful guidelines for future applications of Q-learning to agent-based production scheduling.

Restoration entails the development of a plan consisting of opening or closing of switches, which is called reconfiguration. This dissertation proposes the design of a fast and efficient service restoration with a load shedding method for land-based and ship systems, considering priority of customers and several other system operating constraints. Existing methods, based on centralized restoration schemes that require a powerful central computer, may lead to a single point of failure. This research uses a decentralized scheme based on agents. A group of agents created to realize a specific goal by their interactions is called a Multi-Agent System (MAS). Agents and their behaviors are developed in Java Agent DEvelopment Framework (JADE) and the power system is simulated in the Virtual Test Bed (VTB). The large-scale introduction of Distributed Generators (DGs) in distribution systems has made it increasingly necessary to develop restoration schemes considering DG. The separation of utility causes the system to decompose into electrically isolated islands with generation and load imbalance that can have severe consequences. Automated load shedding schemes are essential for systems with DGs, since the disconnection of the utility can lead to instability much faster than an operator intervention can repair. Load shedding may be the only option to maintain the island when conditions are so severe as to require correction by restoration schemes. Few algorithms have been reported for the problem of maintaining the island, even though load shedding has been reported for power systems using underrequency and under-voltage criteria. This research proposes a new operational strategy for sudden generator-load imbalance due to loss of utility that dynamically calculates the quantity of load to be shed for each island and the quantity of load that can be restored. Results presented in this dissertation are among the first to demonstrate a state-of-the-art MAS for load shedding under islanded conditions and restoration of the shed loads. The load shedding and restoration schemes developed here have behaviors that can incorporate most of the distribution topologies. Achieving service restoration with DG is complicated but new automated switch technologies and communications make MAS a better scheme than existing schemes.

We are recently witnessing an increase in large-scale micro/individual/- granular level behavioural data. Such data has been proven to have the capacity to aid the development of more accurate simulations that will ef- fectively predict the behaviours of complex systems. Despite this increase, the literature has failed to produce a structured modelling approach that will effectively take advantage of such granular data, in modelling com- plex systems that involve social phenomenons (i.e. social complex sys- tems). In this thesis, we intend to bridge this gap by answering the question of how novel structural frameworks, that systematically guides the use of micro-level behaviour and attribute data, directly extracted from the ba- sic entities within a social complex system can be created. These frame- works should involve the systematic processes of using such data to di- rectly model agent attributes, and to create agent behaviour rules, that will directly represent the unique micro entities from which the data was ex- tracted. The objective of the thesis is to define generic frameworks, that would create agent based micro simulations that would directly reflect the target complex system, so that alternative scenarios, that cannot be inves- tigated in the real system, and social policies that need to be investigated before being applied on the social system can be explored. In answering this question, we take advantage of the pros of other model- ing techniques such as micro simulation and agent based techniques in cre- ating models that have a micro-macro link, such that the micro behaviour that causes the macro emergence at the simulation’s global level can be easily investigated. which is a huge advantage in policy testing. We also utilized machine learning in the creation of behavioural rules.This created agent behaviours that were empirically defined. Therefore, this thesis also answers the question of how such structural framework will empirically create agent behaviour rules through machine learning algorithms. In this thesis we proposed two novel frameworks for the creation of more accurate simulations. The concepts within these frameworks were proved using case studies, in which these case studies where from different so- cial complex systems, so as to prove the generic nature of the proposed frameworks. In concluding of this thesis, it was obvious that the questions posed in the first chapter had been answered. The generic frameworks had been created, which bridged the existing gap in the creation of accurate mod- els from the presently available granular attribute and behavioral data, al- lowing the simulations created from these models accurately reflect their target social complex systems from which the data was extracted from.

No description available.

No description available.

No description available.

No description available.

No description available.

No description available.

No description available.