Abstract:

Hierarchical control architectures pose challenges for control, as lower-level dynamics, such as from actuators, are often unknown or uncertain. If not considered correctly in the upper layers, requested and applied control signals will differ. Thus, the actual and the predicted plant behavior will not match, likely resulting in constraint violation and decreased control performance. We propose a model predictive control scheme in which the upper and lower levels - the controller and the actuator - agree on a "contract" that allows to bound the error due to neglected dynamics. The contract allows to guarantee a desired accuracy, enables modularity, and breaks complexity: Components can be exchanged, vendors do not need to provide in-depth insights into the components' working principle, and complexity is reduced, as upper-level controllers do not need full model information of the lower level - the actuators. The approach allows to consider uncertain actuator dynamics with flexible, varying sampling times. We prove repeated feasibility and input-to-state stability and illustrate the scheme in an example for a hierarchical controller/plant cascade.