Model-based knowledge organization: Behaviour preserving model transformations.

摘要

Model Based Knowledge Organization (MBKO) refers to a set of techniques for representing and manipulating knowledge about physical, engineered systems.;The underlying premises of this work are as follows. Many reasoning tasks are dominated by mappings between structural/functional representations of physical systems and behavioural representations. Many different types of structural/functional representations exist, each appropriate to a distinct class of behaviour elicitation task. Often engineering problem solving requires the use of more than one of these representations. Finally, in many engineering problem solving environments a single physical system is persistently the subject of analysis while the specific reasoning tasks are transient and variable in nature.;To support reasoning in such environments, an information system design should focus on the management of structural, functional and behavioural knowledge about the physical system. It should provide the tools for mapping knowledge between modelling domains. Last, it should provide the tools to dynamically compose appropriate models of the physical system on demand.;MBKO provides such a framework. It defines a domain independent language for representing models and behaviour of physical systems (called Model Behaviour Pairs or MBPs). It provides a domain independent means of representing reasoning tasks in terms of incomplete MBPs. It provides a consistent mechanism for translating behavioural knowledge from one model (in one domain) to another model (in another domain). Finally, it provides a set of operators for selecting predictive models which are in some sense minimal and adequate in the context of a specific reasoning task.;The MBKO language and operators are based upon a definition of "dependency" which integrates knowledge about the task, the model topology, the system's behaviour, and the modelling domain. It is this abstraction of domain, system and task knowledge which allows MBKO to consistently handle a diversity of model types.;To support the claim that MBKO is a reasonable representational framework for situated reasoning about physical systems, an implementation has been built and examples run which demonstrate computational improvements over conventional approaches in the modelling domains of planning, qualitative physics and discrete event simulation.