Bernardo Hernandez

Bernardo A. Hernandez, Paul A. Trodden

A new adaptive predictive controller for constrained linear systems is presented. The main feature of the proposed controller is the partition of the input in two components. The first part is used to persistently excite the system, in order to guarantee accurate and convergent parameter estimates in a deterministic framework. An MPC-inspired receding horizon optimization problem is developed to achieve the required excitation in a manner that is optimal for the plant. The remaining control action is employed by a conventional tube MPC controller to regulate the plant in the presence of parametric uncertainty and the excitation generated for estimation purposes. Constraint satisfaction, robust exponential stability, and convergence of the estimates are guaranteed under design conditions mildly more demanding than that of standard MPC implementations.

Bernardo Hernandez, Pablo Baldivieso, Paul Trodden

This paper studies a fundamental relation that exists between stabilizability assumptions usually employed in distributed model predictive control implementations, and the corresponding notions of invariance implicit in such controllers. The relation is made explicit in the form of a theorem that presents sufficient conditions for global stabilizability. It is shown that constraint admissibility of local robust controllers is sufficient for the global closed-loop system to be stable, and how these controllers are related to more complex forms of control such as tube-based distributed model predictive control implementations.

Hugo O. Garcés, Alejandro J. Rojas, Bernardo A. Hernández et al.

This paper compares the performance of model-free controllers on a nonlinear system under cyberattacks, including false data injection and denial-of-service attacks. Four RL reward types are analyzed for accuracy, cost, and resilience. Results show that the Lyapunov reward offers the best resilience with low tracking error. Exponential mode also provides good trade-offs with acceptable resilience under moderate training conditions. Progressive and linear rewards converge faster but are less robust. RL-MPCs show strong steady-state resilience but require longer training times; RL-PID controllers are faster with significantly less training time. Proximal Policy Optimization outperforms Deep Deterministic Policy Gradient with a significant reduction in KPI variance. This study serves to highlight how well-designed RL rewards can improve performance and resilience against cyber threats.

Pablo R Baldivieso Monasterios, Bernardo Hernandez, Paul A Trodden

We propose a distributed model predictive control approach for linear time-invariant systems coupled via dynamics. The proposed approach uses the tube MPC concept for robustness to handle the disturbances induced by mutual interactions between subsystems; however, the main novelty here is to replace the conventional linear disturbance rejection controller with a second MPC controller, as is done in tube-based nonlinear MPC. In the distributed setting, this has the advantages that the disturbance rejection controller is able to consider the plans of neighbours, and the reliance on explicit robust invariant sets is removed.