TL;DR
This paper introduces a novel parallel Model Predictive Control (MPC) algorithm that efficiently handles both state and input constraints for large-scale linear systems. By cleverly designing time-varying constraint margins, the method achieves recursive feasibility and asymptotic stability while maintaining fast, millisecond-range computation times, making it practical for real-time applications.
What Problem Does This Paper Address?
Model Predictive Control (MPC) is a powerful control technique that optimizes future system behavior by repeatedly solving an optimization problem. For large-scale systems, especially those with many states and constraints, solving these optimization problems online can be computationally demanding, hindering real-time application. While parallel MPC schemes exist, they often struggle with the inclusion of state constraints, which are crucial for safety and performance. Existing methods for handling state constraints can lead to conservatism or computational bottlenecks. This paper aims to bridge this gap by developing a parallel MPC approach that effectively incorporates state constraints while guaranteeing recursive feasibility and maintaining fast computation times.
Methods
The core of this work is an extension of a previously proposed parallel MPC scheme. The key innovation lies in the introduction of time-varying, separable constraint margins. These margins are derived using contractive ellipsoidal sets and are designed to ensure that the system remains feasible and stable even when operating with sub-optimal solutions from the parallelized MPC.
Specifically, the authors:
- Develop a strategy for constructing separable time-varying robustness margins for state-constrained linear systems using contractive ellipsoidal sets. This is a crucial step in adapting existing parallel MPC methods to handle state constraints.
- Integrate these time-varying constraints into a real-time parallel MPC framework. This involves modifying the optimization problem to include these new constraints while preserving recursive feasibility and asymptotic closed-loop stability.
- Propose a recursively feasible real-time iteration (RFRTI) variant of the parallel MPC algorithm. This variant replaces the standard optimization step with a modified one that explicitly accounts for the tightened constraints.
- Analyze the trade-off between computational conservatism and online runtime guarantees. The paper shows how tuning parameters can be used to balance the tightness of the constraint margins with the number of iterations required for feasibility.
Datasets
Not discussed in the paper. The paper focuses on a theoretical framework and demonstrates its performance on a simulated large-scale mechatronic system.
Key Insights and Ideas
- Time-varying constraint margins are key to handling state constraints in parallel MPC. By making these margins dependent on time and separable, the algorithm can effectively manage state constraints without sacrificing the parallelizability of the problem.
- Contractive ellipsoidal sets provide a robust mathematical tool for deriving these necessary constraint margins, ensuring stability and feasibility.
- A trade-off exists between conservatism and computational effort. The proposed method allows for tuning to balance how "tight" the constraints are (leading to less conservatism but potentially more computation) versus how many iterations are needed to guarantee feasibility.
- The proposed RFRTI scheme offers a practical path to real-time MPC for complex systems. It provides both performance guarantees and computational efficiency.
Key Results
The paper demonstrates the effectiveness of the proposed RFRTI scheme through a case study of a large-scale mechatronic system with 120 states and 60 control inputs.
- Millisecond-range computation times: The algorithm achieves run-times in the millisecond range, making it suitable for real-time control.
- Near-optimal performance: For a 120-state system with a prediction horizon of 100, the RFRTI controller achieves performance that is only 0.1% suboptimal compared to exact MPC, while being up to 400 times faster.
- Recursive feasibility guarantees: The method provides theoretical guarantees for recursive feasibility, which is a critical aspect for safety-critical applications.
- Negligible performance loss for feasibility: Enforcing recursive feasibility guarantees results in a negligible loss of performance.
Limitations and Future Work
- Conservatism: While the paper addresses conservatism by allowing for tuning, there is an inherent trade-off. The degree of conservatism is controlled by the parameter $r$, which influences the size of the contractive sets.
- Implementation: The paper mentions the implementation in Julia and presents results from a prototype. Future work will focus on an open-source software implementation of the proposed RFRTI control scheme.
- Generalization: The current work focuses on linear systems. Extending the approach to nonlinear systems would be a significant future direction.
Summary updated