Optimization for Control by DTs

A digital twin provides a digital model of a real-world, physical system that is updated in real-time. A motivation for building such a digital representation is being able to control or adapt the physical system. Based on the data flow from the real-world system to the digital twin, real-time values for interesting parameters may be predicted. To take the functionality of a digital twin to the next level, it should also provide a control mechanism: it should create an actionable plan for the next time step.

The plan provided by the control mechanism should be such that, if performed by the real-world system, the result is optimal according to some predefined criteria. The requirements for the digital twin model depend on the control task for which it is intended.

In this project, we focus on this control mechanism and, in particular, on the translation of such a plan to an advice that can be communicated to the user of the digital twin. We investigate the design of such a control mechanism in order to demonstrate its implications for the digital twin modelling phase, which precedes the control phase.

In general, the aim of the control mechanism is to optimize the behavior of the physical system in real-time. This generic real-time optimization problem will take different forms for different digital twins. However, they share issues related to the definition of optimization criteria and goals, the selection of observed, state and control variables, selection of a suitable optimization method, performance, practical applicability, etc.

We will take as a use case the optimization for the control mechanism of the digital twin of the Me, My Diet and I project. A control mechanism for this digital twin will provide personalized dietary advice to the user, with the aim of lowering the long-term postprandial triglyceride response in the user. In this use case, predictions from the digital twin model will be used to automatically generate personalized meal plans.