Abstract
The optimization of process economics within the model predictive control (MPC) formulation has given rise to a new control paradigm known as economic MPC (EMPC). Several authors have discussed the closed-loop properties of EMPC-controlled deterministic systems, however, little have uncertain systems been studied. In this paper we propose EMPC formulations for nonlinear Markovian switching systems which guarantee recursive feasibility, asymptotic performance bounds and constrained mean square (MS) stability.
Introduction
Traditional Model Predictive Control (MPC) is typically designed to drive a system toward a specific set-point or trajectory. In contrast, Economic MPC (EMPC) optimizes a performance metric directly—such as maximizing profit or minimizing energy consumption—without requiring a predefined steady state. While EMPC is well-established for deterministic systems, real-world engineering often involves significant uncertainty.
Many industrial systems, such as power grids or chemical reactors, switch between different operating modes. These are modeled as Markovian switching systems. This paper addresses a critical theoretical gap: how to ensure that an economically optimized controller remains stable and feasible when the system's underlying dynamics change stochastically.
Problem Formulation
The primary challenge in stochastic EMPC is ensuring that the optimization remains solvable (feasible) over time. In a switching environment, an action that is economically optimal for the current mode might lead the system into a state that is impossible to recover from if the mode changes unexpectedly. Furthermore, standard stability proofs for tracking-based MPC do not naturally apply to economic costs, which may not have a minimum at the desired operating point.
Proposed Methodology
The authors propose two primary EMPC formulations to handle nonlinear Markovian switching:
1. Terminal Equality Constraints and the "Bet Node"
To maintain feasibility across mode transitions, the authors introduce the "bet node" strategy. The controller identifies the most likely successor mode in the Markov chain. By forcing the predicted terminal state to match the optimal steady state of this "bet node," the controller ensures it is always prepared for the most probable future transition, balancing economic gain with operational safety.
2. Terminal Regions and Penalty Functions
For systems requiring more flexibility, the authors utilize terminal sets ($X_f$) and penalty functions ($V_f$). These are designed using local linearization and Bilinear Matrix Inequalities (BMIs). This approach allows the controller to operate within a region rather than a single point, providing a buffer against stochastic fluctuations.
3. Stochastic Dissipativity and Cost Rotation
To prove stability, the authors employ Stochastic Dissipativity. They use a "rotated cost function" technique to mathematically transform the economic cost into a form that behaves like a traditional tracking cost. This allows for the application of Lyapunov-style arguments to prove stability without changing the actual control actions or the economic objective.
Key Theoretical Results
- Recursive Feasibility: The paper provides a rigorous mathematical proof that if the EMPC problem is feasible at the initial time step, it will remain feasible for all future steps, despite random mode switching.
- Mean Square Stability (MSS): The authors establish sufficient conditions for MSS, ensuring that the system state converges in a statistical sense and does not diverge due to economic optimization.
- Performance Guarantees: The framework guarantees that the asymptotic average performance of the system is at least as good as the best possible steady-state operation, providing a formal justification for the economic approach.
Design Framework for Engineers
The paper offers a concrete computational path for implementation. By solving a set of matrix inequalities, engineers can determine the necessary terminal costs and regions to guarantee a stable, economically optimal controller. While the core theory addresses nonlinear systems, the authors provide a bridge for practical application via local linear approximations.
Limitations and Future Work
- Conservatism: The "bet node" strategy may be overly cautious if the Markov transition probabilities are widely and evenly distributed.
- Local vs. Global Stability: Because the design guidelines for terminal regions rely on local linearization, the stability guarantees are primarily valid within a neighborhood of the optimal equilibrium.
- Computational Complexity: Solving BMIs is more computationally intensive than standard Linear Matrix Inequalities (LMIs), which may present a bottleneck for high-dimensional systems.
Future research directions include relaxing the assumption of a common equilibrium across all modes to better accommodate systems with vastly different operational requirements in different phases.
Summary updated