A methodology to design rule-based systems.

摘要

Accurate and responsive systems for forecasting and operational control are crucial in making an enterprise successful in today's highly competitive market. Conventional mathematical modeling approaches that are utilized in a forecasting and operational control system are not adequate in handling unexpected market and environmental circumstances. Rule-based approaches, where the enterprise strategies are written as IF-THEN rules, provide timely access for enterprise managers to modify their forecasting and operational control systems directly using their expertise, and thus increase the forecasting accuracy and the operational control system's responsiveness. Rule-based systems have been shown to be successful in many applications such as automatic control, system monitoring and technical diagnosis. Although the rule-based paradigm seems relatively conceptually simple, in realistic systems it is difficult to design and implement a rule-based system that works in a correct way. One challenge to designing a rule-based system is that the rules acquired from the enterprise experts may be incomplete, ambiguous and even conflicting.;In this dissertation, we have developed a new methodology for rule-based system design. We start with a set of rules suggested by enterprise experts and prove they can be represented as a finite state automaton (a Moore machine). This provides the dynamics of the system evolution as described by rules. We formulate an optimal control problem and treat the rules as binary decision variables. The solution to the optimal control problem is transformed back to the rule-space to obtain an optimal set of rules that constitute the rule-based system to be designed.;We demonstrate our methodology in the applications of supply chain management and power system operations with smart grids.