HIERARCHICAL-GRANULARITY HOLONIC MODELLING

This thesis aims to introduce an agent-based system engineering approach, named Hierarchical-Granularity Holonic Modelling, to support intelligent information processing at multiple granularity levels. The focus is especially on complex hierarchical systems. Nowadays, due to ever growing complexity of information systems and processes, there is an increasing need of a simple self-modular computational model able to manage data and perform information granulation at different resolutions (i.e., both spatial and temporal). The current literature lacks to provide such a methodology. To cite a relevant example, the object-oriented paradigm is suitable for describing a system at a given representation level; notwithstanding, further design effort is needed if a more synthetical of more analytical view of the same system is required. In the literature, the agent paradigm represents a viable solution in complex systems modelling; in particular, Multi-Agent Systems have been applied with success in a countless variety of distributed intelligence settings. Current agent-oriented implementations however suffer from an apparent dichotomy between agents as intelligent entities and agents’ structures as superimposed hierarchies of roles within a given organization. The agents’ architectures are often rigid and require intense re-engineering when the underpinning ontology is updated to cast new design criteria. The latest stage in the evolution of modelling frameworks is represented by Holonic Systems, based on the notion of ‘holon’ and ‘holarchy’ (i.e., hierarchy of holons). A holon, just like an agent, is an intelligent entity able to interact with the environment and to take decisions to solve a specific problem. Contrarily to agent, holon has the noteworthy property of playing the role of a whole and a part at the same time. This reflects at the organizational level: holarchy functions first as autonomous wholes in supra-ordination to their parts, secondly as dependent parts in sub-ordination to controls on higher levels, and thirdly in coordination with their local environment. These ideas were originally devised by Arthur Koestler in 1967. Since then, Holonic Systems have gained more and more credit in various fields such as Biology, Ecology, Theory of Emergence and Intelligent Manufacturing. Notwithstanding, with respect to these disciplines, fewer works on Holonic Systems can be found in the general framework of Artificial and Computational Intelligence. Moreover, the distance between theoretic models and actual implementation is still wide open. In this thesis, starting from the Koestler’s original idea, we devise a novel agent-inspired model that merges intelligence with the holonic structure at multiple hierarchical-granularity levels. This is made possible thanks to a rule-based knowledge recursive representation, which allows the holonic agent to carry out both operating and learning tasks in a hierarchy of granularity levels. The proposed model can be directly used in terms of hardware/software applications. This endows systems and software engineers with a modular and scalable approach when dealing with complex hierarchical systems. In order to support our claims, exemplar experiments of our proposal are shown and prospective implications are commented.