Hybrid Data-enabled Predictive Control: Incorporating model knowledge into the DeePC

TL;DR

This paper introduces Hybrid Data-enabled Predictive Control (HDeePC), which integrates partial model knowledge into data-driven predictive control, enhancing robustness and computational efficiency for linear, time-varying, and nonlinear systems.

Contribution

It develops a novel hybrid control approach combining data-driven and model-based methods, with proven theoretical properties and practical case studies demonstrating its effectiveness.

Findings

Feasible set equivalence in noiseless LTI systems

Improved robustness to noise compared to pure data-driven methods

Successful application to microgrid energy storage and power systems

Abstract

Predictive control can either be data-based (e.g. data-enabled predictive control, or DeePC) or model-based (model predictive control). In this paper we aim to bridge the gap between the two by investigating the case where only a partial model is available, i.e. incorporating model knowledge into DeePC. In our formulation, the partial knowledge takes the form of known state and output equations that are a subset of the complete model equations. We formulate an approach to take advantage of partial model knowledge which we call hybrid data-enabled predictive control (HDeePC). We prove feasible set equivalence and equivalent closed-loop behavior in the noiseless, LTI case. As we show, this has potential advantages over a purely data-based approach in terms of computational expense and robustness to noise in some cases. Furthermore, this allows applications to certain linear time-varying and nonlinear systems. Finally, a number of case studies, including the control of an energy storage system in a microgrid, a triple-mass system, and a larger power system, illustrate the potential of HDeePC.