Adaptation strategies for self-organising electronic institutions

Abstract

For large-scale systems and networks embedded in highly dynamic, volatile, and unpredictable environments, self-adaptive and self-organising (SASO) algorithms have been proposed as solutions to the problems introduced by this dynamism, volatility, and unpredictability. In open systems it cannot be guaranteed that an adaptive mechanism that works well in isolation will work well — or at all — in combination with others. In complexity science the emergence of systemic, or macro-level, properties from individual, or micro-level, interactions is addressed through mathematical modelling and simulation. Intermediate meso-level structuration has been proposed as a method for controlling the macro-level system outcomes, through the study of how the application of certain policies, or norms, can affect adaptation and organisation at various levels of the system. In this context, this thesis describes the specification and implementation of an adaptive affective anticipatory agent model for the individual micro level, and a self-organising distributed institutional consensus algorithm for the group meso level. Situated in an intelligent transportation system, the agent model represents an adaptive decision-making system for safe driving, and the consensus algorithm allows the vehicles to self-organise agreement on values necessary for the maintenance of “platoons” of vehicles travelling down a motorway. Experiments were performed using each mechanism in isolation to demonstrate its effectiveness. A computational testbed has been built on a multi-agent simulator to examine the interaction between the two given adaptation mechanisms. Experiments involving various differing combinations of the mechanisms are performed, and the effect of these combinations on the macro-level system properties is measured. Both beneficial and pernicious interactions are observed; the experimental results are analysed in an attempt to understand these interactions. The analysis is performed through a formalism which enables the causes for the various interactions to be understood. The formalism takes into account the methods by which the SASO mechanisms are composed, at what level of the system they operate, on which parts of the system they operate, and how they interact with the population of the system. It is suggested that this formalism could serve as the starting point for an analytic method and experimental tools for a future systems theory of adaptation.