The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #9720

Title:Assumption-Based Reasoning for Dynamic Model Selection: Implementing the Graph of Models
Authors:Da Silva,LP
Date:Sep 1997
Presented:
Keywords:
Abstract:In order to cope with a real system's complexities and the huge amount of knowledge needed to describe the reason about its behaviour, models are used as abstractions that capture only the system's relevant characteristics for a given level of analysis. When the level of detail of the model does not satisfy the current requirements, a new model of the system, with a different level of description is needed. There are two main approaches to reasoning about multiple models: model composition and model selection. In model selection a model is selected from a predefined candidate set of models, in which all models represent the same object. Conversely, in model composition a model is built by automated selection of a set of model fragments. The Graph of Models [Addanki et al. 91] is a paradigm that implements model selection in which a graph is used to represent the candidate set of models. The project consists of implementing the Graph of Models technique and testing its capabilities as a tool that automatically finds which level of detail or particular view of a given system can describe the current state of affairs. In this approach, a model has an associated list of supporting assumptions. Each node in the graph represents a model and the models themselves are stored in a model-base. the assumption transitions are used to represent the connections between nodes. Once a conflict is found between the current model's predictions and the real system's measurements, the graph tries to find the simplest model that best describes the real system's behaviour. This implementation was successfully tested using two simple systems from the domains of Chemical Engineering and Physics.
Download:NO ONLINE COPY

[Search These Pages] [DAI Home Page] [Comment]