A Flexible Move Blocking Strategy to Speed up Model-Predictive Control while Retaining a High Tracking Performance

This paper presents a strategy to reduce the
complexity and thus the computational burden in modelpredictive
control (MPC) by a flexible online input move
blocking algorithm. Model-predictive sampled-data control of
constrained, LTI plants is considered. Move blocking is an input
parameterisation in MPC where the control input is forced
to be constant over several prediction sample steps to reduce
the dimensionality of the underlying optimisation problem.
Typically, the prediction sample steps where the control input is
not allowed to vary (i. e. the block distribution) is predetermined
offline and is kept constant throughout the control operation.
However, the control performance and computational efficiency
can be improved if the block length is adjusted to the specific
operating conditions. In this work, a heuristic method to adjust
the block length online according to the initial state of the system,
reference signals, measured disturbances and constraints
is presented. A numerical example shows the effectiveness of
the approach.