Linear Data-Driven Economic MPC with Generalized Terminal Constraint

Yifan Xie, Julian Berberich, Frank Allgöwer

Abstract

In this paper, we propose a data-driven economic model predictive control (EMPC) scheme with generalized terminal constraint to control an unknown linear time-invariant system. Our scheme is based on the Fundamental Lemma to predict future system trajectories using a persistently exciting input-output trajectory. The control objective is to minimize an economic cost objective. By employing a generalized terminal constraint with artificial equilibrium, the scheme does not require prior knowledge of the optimal equilibrium. We prove that the asymptotic average performance of the closed-loop system can be made arbitrarily close to that of the optimal equilibrium. Moreover, we extend our results to the case of an unknown linear stage cost function, where the Fundamental lemma is used to predict the stage cost directly. The effectiveness of the proposed scheme is shown by a numerical example.

Introduction

Traditional control strategies typically focus on tracking a specific setpoint. In contrast, Economic Model Predictive Control (EMPC) aims to optimize a general performance metric, such as minimizing energy consumption or maximizing yield, where the optimal operating point (equilibrium) may not be known a priori.

Most existing EMPC strategies rely on two critical pieces of information that are often unavailable in practice: an accurate physics-based model of the system and prior knowledge of the optimal steady state. While data-driven methods have emerged to bypass model identification, they often still require a predefined target setpoint. This paper introduces a data-driven EMPC scheme that optimizes unknown linear systems without requiring a model or prior knowledge of the optimal operating point, allowing the system to autonomously discover and converge to an economically optimal steady state.

Methodology

The proposed framework integrates several key concepts to achieve model-free economic optimization:

  • Willems’ Fundamental Lemma: This principle allows for the prediction of future system trajectories using a "persistently exciting" historical input-output data sequence. By representing any future trajectory as a linear combination of collected data, the need for an explicit state-space model ($Ax + Bu$) is eliminated.
  • Generalized Terminal Constraints & Artificial Equilibrium: Instead of forcing the controller toward a fixed terminal point, the authors employ a generalized constraint. The system is required to reach some valid equilibrium at the end of the prediction horizon. This equilibrium is treated as a decision variable (an "artificial equilibrium"), allowing the optimization algorithm to determine the most cost-effective steady state dynamically.
  • Penalty Parameters: A penalty term ($\beta$) is added to the cost function to ensure stability and proximity to the artificial equilibrium. The authors prove that as $\beta$ increases, the system's performance approaches the theoretical optimum.
  • Handling Unknown Costs: For scenarios where the economic cost function parameters are unknown, the authors extend the Fundamental Lemma to predict cost values directly from data, treating the stage cost as an additional system output.

Experimental Evaluation

The approach was validated using a numerical simulation of a linearized chemical reactor, a standard benchmark in EMPC literature. The system consists of 3 states, 1 input, and 2 outputs. An input-output trajectory of length $N=100$ was generated using random inputs to serve as the initial data set.

Convergence of the stage cost to the optimal value for different penalty parameters.

As shown in the figure above, the data-driven controller successfully drives the system to the optimal stage cost of approximately -0.6396. Notably, the choice of the penalty weight ($\beta$) affects the transient behavior but does not prevent convergence to the optimum, demonstrating the robustness of the scheme. Unlike previous methods, this controller reached the optimal point without being provided with its coordinates at initialization.

Key Insights and Results

  • Model-Free Optimality: The controller performs exploration and optimization simultaneously, eliminating the need for offline model identification or steady-state analysis.
  • Recursive Feasibility: The authors provide mathematical proof that if the optimization problem is solvable at the initial time step, it remains solvable for all future steps, ensuring continuous operation.
  • Performance Guarantees: The "asymptotic average performance" can be made arbitrarily close to the theoretical optimum by tuning the penalty parameter $\beta$.

Limitations and Future Work

While promising, the current theory is limited to Linear Time-Invariant (LTI) systems. Many real-world applications, such as robotics or complex chemical processes, exhibit high nonlinearity. Additionally, the current proofs assume "clean" data; extending these guarantees to handle measurement noise in the economic context is a primary direction for future research. Finally, the "unknown cost" feature currently requires a linear cost function; extending this to quadratic or general convex costs would significantly increase the method's versatility.

Summary updated

Abstract

In this paper, we propose a data-driven economic model predictive control (EMPC) scheme with generalized terminal constraint to control an unknown linear time-invariant system. Our scheme is based on the Fundamental Lemma to predict future system trajectories using a persistently exciting input-output trajectory. The control objective is to minimize an economic cost objective. By employing a generalized terminal constraint with artificial equilibrium, the scheme does not require prior knowledge of the optimal equilibrium. We prove that the asymptotic average performance of the closed-loop system can be made arbitrarily close to that of the optimal equilibrium. Moreover, we extend our results to the case of an unknown linear stage cost function, where the Fundamental lemma is used to predict the stage cost directly. The effectiveness of the proposed scheme is shown by a numerical example.
BibTeX
@inproceedings{xie2023linear,
  author = {Yifan Xie and Julian Berberich and Frank Allgöwer},
  title = {{Linear Data-Driven Economic MPC with Generalized Terminal Constraint}},
  booktitle = {IFAC-PapersOnLine},
  year = {2023},
  doi = {10.1016/j.ifacol.2023.10.209},
}
Figure