As of today, the 1st edition of the Handbook of Digital Twins (CRC Press / Taylor&Francis Group) is out, featuring a chapter on “Automated Inference of Simulators in Digital Twins“, co-authored with Eugene Syriani.
Preprint available.
In this chapter, we introduce the reader to automated simulator construction by machine learning. We rely on the Discrete Event System Specification (DEVS) as the simulation formalism, and choose reinforcement learning as the machine learning method for inference. The reinforcement learning agent learns simulator components through virtual experimentation and gradual adaptation of the simulation model. The approach leverages the unique architectural opportunities of digital twins, particularly, the availability of advanced instrumentation (via sensors and actuators) on the digital side.
Through automation, simulators can be developed more efficiently, economically, and their maintenance is simplified as well. We demonstrate the approach through a case of a complex cyber-biophysical system.
