Stability-informed Bayesian Optimization for MPC Cost Function Learning
This work addresses the problem of learning MPC parameters safely and stably for control systems, representing an incremental improvement by adding stability constraints to an existing Bayesian optimization framework.
The paper tackles the challenge of designing predictive controllers for optimal closed-loop performance while ensuring safety and stability under imperfect information, by using constrained Bayesian optimization to learn a neural network-based cost function for model predictive control (MPC) with stability constraints derived from the MPC's optimal value function as a Lyapunov candidate. The result is demonstrated in simulations, showing improved performance and safety capabilities.
Designing predictive controllers towards optimal closed-loop performance while maintaining safety and stability is challenging. This work explores closed-loop learning for predictive control parameters under imperfect information while considering closed-loop stability. We employ constrained Bayesian optimization to learn a model predictive controller's (MPC) cost function parametrized as a feedforward neural network, optimizing closed-loop behavior as well as minimizing model-plant mismatch. Doing so offers a high degree of freedom and, thus, the opportunity for efficient and global optimization towards the desired and optimal closed-loop behavior. We extend this framework by stability constraints on the learned controller parameters, exploiting the optimal value function of the underlying MPC as a Lyapunov candidate. The effectiveness of the proposed approach is underlined in simulations, highlighting its performance and safety capabilities.