CODEVS: AN EXTENSION OF DEVS FOR INTEGRATION OF SIMULATION AND MACHINE LEARNING

Abstract

When we model a system to analyse it, there are two main methods we can use. First, there are knowledge-based simulation modelling methods using system operations, such as discrete event system specification (DEVS). Conversely, there are data-driven modelling methods using data analysis without explicit system knowledge, such as machine learning. These two models can be used appropriately in situations where it is difficult to model sufficiently with one method, and through this, the advantages of each method can be maximised. In other words, for this, a method is required to specify one system by using two methods at the same time. Therefore, in this paper, we introduce an extension of DEVS formalism, called Cooperative DEVS (CoDEVS), which enables representation of both a simulation model and a machine learning model. It consists of a simulation model, data model, and interface models that convert events between the simulation and data models. We also introduce a modified simulation algorithm that can interpret the new formalism and simulate a distributed file system to show the validity of the proposed work.