A Modularization Approach for Nonlinear Model Predictive Control of Distributed Fast Systems

Distributed interconnected systems are omnipresent today.
The development of advanced control methods for such
systems are still challenging. Herein, the real-time
applicability, flexibility, portability and ease of
implementation are issues of the existing control
solutions, especially for more advanced methods such as
model predictive control. This paper is addressing these
issues by presenting an efficient modular composition
scheme for distributed fast nonlinear systems. The
advantage of this modularization approach is the capability
of changing control objectives, constraints, dynamics and
system topology online while maintaining fast computation.
This work analyzes the functions that have to be provided
for a continuation generalized minimal residual method
(CGMRES) model predictive controller based on the
underlying control problem. The specific structure of these
functions allows their decomposition into suitable fast
modules. These modules are then used to recompose the
functions which are required for the control of distributed
systems in a computational efficient way, while maintaining
the flexibility to dynamically exchange system parts. To
validate this computational efficiency, the computation
time of the proposed modular control approach is compared
with a standard nonmodular implementation in a pursuit
scenario of quadrotor unmanned aerial vehicles (UAV).
Furthermore the real-time applicability is discussed for
the given scenario.