text
stringlengths
0
89.3k
x0xk 4a
xi1 A Axi B Buiwi 4b
xi XAB Dwi W 4c
ui U 4d
xN Sf 4e
where xiui and widenote the predicted state input
and disturbance respectively and Sf X is a robust
positively invariant RPI set according to Definition 31
Definition 3 RPI set for 1The set Sf X is a
robust positively invariant RPI set for system 1with
control law uKfx Ufor all x Sf ifx Sf
x Sffor all w WAB D
In this paper we solve the CSP using a tubebased MPC
formulation 313 that tightens the constraints with so
called tubes for a shorter horizon N Nand apply the
MPC in shrinking horizon or receding horizon In the
1Note that we use an RPI set for simplicity however a
robust control invariant RCI set could be used insteadfollowing section we detail how to derive such a formu
lation by generalizing the tubebased MPC formulation
proposed in 25
3 Filterbased System Level TubeMPC
The filterbased SLTMPC proposed in 25 relies on de
signing an auxiliary disturbance set W before trans
forming it online to overapproximate the combined ef
fect of additive and parametric uncertainties This en
ables online adaptation of the tube controller and re
sults in a nonconservative tubebased MPC method We
extend this method to general polytopic disturbance
sets WwHwwhwinstead of normbounded
disturbance sets Ww w1 which extends
the scope of the method and simplifies its exposition
We first rewrite the prediction dynamics 4b by collect
ing all uncertainties in a separate term denoted as ηi
xi1AxiBuiηi 5
ηi Axi Buiwi 6
Then following the idea presented in 25 we over
approximate the combined uncertainty ηifori
0 N1 with disturbance tubes ie
ηi pi F iΣpi iM
j0Σi1jW7
where piRn Σi1jRnnare online optimization
variables denoting the nominal disturbance and the dis
turbance filter respectively The compact polytopic dis
turbance set W Rn which contains the origin in its
interior is designed offline Note that these disturbance
tubes 7 are a generalization of homothetic tubes 8
This can be seen by setting Σ i1iαiInand Σ i1j 0
forj 0 i1 which simplifies the containment con
dition 7 to ηi pi αiW where αiare the dilation
factors and piare the tube centers in 8 If we enforce 7
for all w W AB D during online optimiza
tion we can guarantee that for optimized piΣi1j
there always exists a sequence of wjW j 0 i
such that
ηipiiX
j0Σi1jwj i 0 N1
Therefore we can equivalently write dynamics 5 as
xi1AxiBuipiiX
j0Σi1jwj 8
when imposing that 7 holds In the remainder of the
paper we use the auxiliary dynamics 8 and formulate
3the filterbased SLTMPC problem by splitting 8 into
nominal and error dynamics We use ziandeias the
nominal and the error state respectively such that xi
ziei Together with the tubebased control policy ui
viνi where viandνiare the nominal and error inputs
we obtain
xi1Aziei Bviνi piiX
j0Σi1jwj
and thus as is standard in tubebased MPC 3 sep
aration into nominal and error dynamics ie
zi1AziBvipi 9
ei1AeiBνiiX
j0Σi1jwj 10
Since we consider MPC with horizon length N we define
the trajectories z z
0 z
Nv v
0 v
N
p p
0 p
N1w w
0w
N1e
e
1 e
N and ν ν
1 ν
Nas the nominal
state nominal input nominal disturbance auxiliary
disturbance error state2and error input trajectories
respectively The nominal and error dynamics are then
compactly written in terms of these trajectories as
zZAN1zZBN1v
z0
p
11
eZANeZBNνΣw 12
where Zis a lower shift matrix ie a matrix with ones