DEVS execution acceleration with machine learning
Discrete Event System Specification DEVS separates modeling and simulation execution. Simulation execution is done within a runtime environment that is often called a DEVS simulator. This separation creates an opportunity to incorporate smart algorithms in the simulator to optimize simulation execution. In this paper, we propose incorporating some predictive machine learning algorithms into the DEVS simulator that can cut simulation execution times significantly for many simulation applications without compromising the simulation accuracy. In this paper, we introduce a specific learning mechanism that can be embedded into the DEVS simulator to incrementally build a predictive model that learns from past simulations. We further look into issues related to the predictive model selection, its prediction accuracy, its effect on the overall simulation performance, and when to switch between the predictive model and the simulation during an execution.
|Discrete Event System Specification DEVS, Machine learning, Regression models|
|2016 TMS/DEVS Symposium on Theory of Modeling and Simulation, TMS/DEVS 2016, Part of the 2016 Spring Simulation Multiconference, SpringSim 2016|
|Organisation||Department of Systems and Computer Engineering|
Saadawi, H. (Hesham), Wainer, G.A, & Pliego, G. (German). (2016). DEVS execution acceleration with machine learning. In Proceedings of the 2016 Spring Simulation Multiconference - TMS/DEVS Symposium on Theory of Modeling and Simulation, TMS/DEVS 2016.