Development of a knowledge-assisted linear programming modeling system

Abstract

Linear Programming (LP) models have been used to support managerial decision-makings in various domains. However, the knowledge-intensive nature of model formulation and the need of model maintenance have inhibited their use by non-expert users. To overcome these difficulties, aiding the modeling process of LP models has attracted a great deal of research in model management area. In this line of research, we propose a framework for the knowledge-assisted LP modeling system. Our major research goal is to assist decision makers who may not be experts in modeling so that they can easily formulate and maintain LP models without any ambiguities and errors. To identify the research targets, we first articulate the design criteria of a knowledge-assisted modeling system. The most distinctive criterion pursued in this thesis is the independent management of knowledge base of multiple LP models and data. To achieve these criteria, we suggest four views of model representation: semantic view, modeling language view, mathematical notational view, and tabular view. In semantic view, the associations among attributes, BOT``s and constraints are represented in a constraint network. We develop the formulation reasoning process which constructs a specific semantic LP model using the common modeling knowledge base (CMKB). CMKB comprises the modeling structure knowledge and domain knowledge. Modeling structure knowledge is used to identify the associations among modeling constructs, while domain knowledge is used to interactively assist the non-expert user in specifying the problem. When the CMKB is changed, multiple specific models formulated using the old knowledge base should be changed to preserve their consistencies with the new knowledge base. For this purpose, we develop the change propagation process. For knowledge acquisition and maintenance, we apply the case-based learning approach. In our approach, an important underlying assumption is that modeling experts and know...