Model-based knowledge organization: A framework for constructing high-level control systems

摘要

This paper presents a framework for the representation of knowledge about physical systems that is tailored for use in the development of high-level control systems for large industrial processes (where high-level control tasks include those activities such as planning, diagnosis, and plant-wide process optimization, which are not easily automated using conventional techniques). It is intended to streamline the development of such systems, and to improve their maintainability and extensibility (in the same manner as data dictionaries/database scemata have streamlined conventional software engineering).The salient features of the representation scheme presented here are: It isolates application-specific knowledge from more general knowledge (through the concepts of models and modeling domains); it provides specific structures for representing behavioural information derived either directly from the outside world or from model-based experiments (e.g., simulation); and it provides a means of representing plans or sequences of action (for planning of activitieswithin the computer, and for plans concerning the outside world). This framwork is designed to be extensible: A new analytical technique may be implemented under an existing modeling domain or a new domain may be defined and integrated. It is also intended to be platform-independent: the representation framework exists on both the development and runtime platforms, and so a bridge exists for porting knowledge from one to the other.