text
stringlengths
0
89.3k
lem as shown in Appendix A1
Remark 1 Note that we can relax the structural con
straint Σi1iσIntoΣi1idiagσi1 σin if
we design Was a hyperrectangle therefore recovering the
same disturbance tube parameterization as in 25 For
more details see Appendix A1
Finally we use nominal dynamics 9 error dynam
ics 15 and disturbance overapproximation 22 to
5formulate the generalized version of the filterbased
SLTMPC problem 25 as
min
zvpΣ
ΦeΦνlfzN N1X
i0lzi vi 23a
sti 0 N1
z0xk 23b
zi1AziBvipi 23c
h
INnZANZBNi
Φe
Φν
Σ 23d
zi X F iΦe 23e
vi U F iΦν 23f
zN Sf FNΦe 23g
ψd
ii1M
j0Ψd
ijWW σi1W d 1 nD
23h
where l and lf are suitable stage and terminal
costs ψd
iand Ψd
ijare defined as in 18 Sfis an RPI ter
minal set according to Definition 3 and FiΦeFiΦν
are the state and input tubes defined as
FiΦei1M
j0Φe
ijWFiΦνi1M
j0Φν
ijW24
In order to show recursive feasibility of 23 we would
need to show that Sf F NΦe is RPI itself How
ever since Sfis RPI with respect to both w W
and AB D Definition 3 this is difficult be
causeFNΦe would need to be an exact reachable set
of 1 for all w W AB D This is clearly
not the case since FNΦe is computed via 17 which
is an overapproximation of the combined uncertainties
Therefore we need to restrict the SLTMPC 23 to a
shrinking horizon regime similar to 1625 which switch
between solving the MPC problem with a shrinking hori
zon and exactly solving the robust CSP 4 for N 1
to show recursive feasibility and robust stability How
ever this strategy requires implementation of a switch
ing logic and only works well if task horizon Nis finite
and known in advance In the next section we show how
to modify the terminal constraints in 23 such that the
resulting MPC can be applied in receding horizon For
a recursive feasibility proof of 23 in shrinking horizon
we refer to 16 Appendix 6
4 Recursively Feasible Filterbased SLTMPC
The terminal constraints in 23 are not suitable to prove
recursive feasibility in receding horizon due to Sfbeing
computed for system 1 with combined uncertainty ηandFNΦe being computed for auxiliary system 8
with only additive uncertainty w Therefore we propose
a new set of terminal constraints that only rely on sets
computed for auxiliary system 8 The key idea is to
exploit the separation xzeand formulate a sepa
rate terminal control law for both the nominal and error
states ensuring that both only depend on w The result
ing new terminal set Xfis then used to constrain the
terminal state xN Xfin the proposed MPC scheme
For this we first define an auxiliary RPI set Zffor a sim
plified version of 8 with pi 0 Σ i1iInΣi1j 0
for all iandj 0 i1 ie
xk 1 Axk Buk wk 25
Definition 4 RPI set for 25The set Zf X is a
robust positively invariant RPI set for system 25with
control law uKfx Ufor all x Zf ifx Zf
x Zffor all wkW
We then use the control law Kfof RPI set Zfto con
struct the terminal control law as
κfxκz
fzκe
feKfzνKfzNX
j0Φν
NjwNj
26
where Φν
Nj ie the last block row of Φν can be freely
optimized since it is not constrained by 15 Before con
structing the RPI set corresponding to control law 26
we first overapproximate the terminal uncertainty ηN
with a terminal disturbance tube similar to 74ie
ηN FNΞN1M
j0ΞjW σNW 27
where Ξ Ξ 0 Ξ N1 are additional disturbance fil
ter parameters Similar to Section 3 there always ex
ists a sequence of wjW j 0N such that ηNPN1