Oracle-based economic predictive control

TL;DR

This paper introduces an economic model predictive control approach using an oracle prediction model for the economic cost, ensuring stability and robustness even with imperfect predictions, demonstrated on a chemical reactor case study.

Contribution

It proposes a novel oracle-based economic predictive control framework with stability guarantees and practical robustness for nonlinear systems.

Findings

Oracle can be derived for general nonlinear systems with Morse economic cost functions.

Stability of the controller is proven under nominal and practical conditions.

Case study demonstrates effectiveness in a chemical reactor process.

Abstract

This paper presents an economic model predictive controller, under the assumption that the only measurable signal of the plant is the economic cost to be minimized. In order to forecast the evolution of this economic cost for a given input trajectory, a prediction model with a NARX structure, the so-called oracle, is proposed. Sufficient conditions to ensure the existence of such oracle are studied, proving that it can be derived for a general nonlinear system if the economic cost function is a Morse function. Based on this oracle, economic model predictive controllers are proposed, and their stability is demonstrated in nominal conditions under a standard dissipativity assumption. The viability of these controllers in practical settings (where the oracle may provide imperfect predictions for generic inputs) is proven by means of input-to-state stability. These properties have been illustrated in a case study based on a continuously stirred tank reactor.