Model specification and analysis for discrete event simulation

TR Number
Date
1982
Journal Title
Journal ISSN
Volume Title
Publisher
Virginia Polytechnic Institute and State University
Abstract

Several authors assert that any significant improvement in the efficiency in the development and utilization of simulation models requires the assistance of a Model Management System to automate much of the process. This work develops one aspect of such system: tools fundamental to the specification and analysis of discrete event models. A model specification language is developed which allows analysis of a model specification as the specification is being developed. Model analyses are discussed which (1) detect several types of errors in a model specification, (2) automate the generation of many types of documentation useful during model development and the generation of model documentation useful during the model life cycle, (3) improve model implementations by assisting in the choice of a world view and implementation technique.

A model specification in this language is called a condition specification (CS). Formal procedures are developed to transform a cs into a model specification in any of the three traditional worldviews of event scheduling, activity scanning, and process interaction. These transformation procedures are supported by a careful definition of the equivalency of model specifications and by a characterization of each of the three world views. Results are proved which show that the approach of each worldview transformation is valid (i.e., result in equivalent model specifications). This is necessary since the transformation procedures may delete parts of a CS. Both the characterization of each world view and the procedures that transform a cs into each world view provide a better understanding of the nature of each worldview than has existed to date.

Several necessary properties for error-free model specifications are identified and defined. While useful test procedures for these properties can be developed, most of these properties are shown to be unverifiable in an absolute sense. That is, it is proved that no algorithm is possible which can detect every instance of several important model specification errors.

Description
Keywords
Citation