(An) approach to solver integration for modeling environment

Abstract

One of the essential issues in Model Management System(MMS) is how to facilitate good management of key resources, namely data, model, solvers, and knowledge. Current MMSs address the problems of managing the key resources. However, little efforts can be found for the management of solvers. To manage solvers like model and data as a key resources of organization, solvers must be integrated with modeling environment in a formal way. In this paper, an extensible solver integration approach is proposed. The main idea is to separate the embedded solver-related knowledge from implementation codes for data structures, input descriptions, output handlings, etc. In this thesis, the extensible solver integration scheme is suggested which includes the constructs of data structures, solver input file structure description language, and output manipulation language. Integration of solvers for optimization is used as an illustration example. The extensible solver integration approach makes it possible to control the interaction between modeling system and solver, and to add a new solver into modeling environment. It may leads to a way for solver base construction and management of solvers and related knowledge.