Datasets:
tensorkelechi
/
arxiv-cstext

Dataset card Data Studio Files Community
1
text
stringlengths
0
89.3k
5 SLTMPC with Asynchronous Updates
Both the proposed filterbased SLTMPC scheme 32
and the method presented in 25 are computationally
demanding due to the concurrent optimization of nom
inal trajectories zvp error dynamics 15 parame
terized by error system responses ΦeΦνand distur
bance filter ΣΞ and terminal set scaling α This ren
ders both methods unsuitable for practical applications
with fast sampling times and limited computational re
sources To alleviate some of the computational burden
11we extend the asynchronous computation scheme intro
duced in 28 for the proposed filterbased SLTMPC To
reduce the complexity of 32 we remove error dynam
ics parameters ΦeΦνΣΞ and terminal scaling α
as optimization variables and instead use values that
are computed at a lower frequency The asynchronous
computation scheme uses this idea to split the opti
mization of 32 into two separate processes the pri
mary and the secondary process The secondary pro
cess computes the error dynamics parameters and the
terminal set scaling before passing them to the pri
mary process which only optimizes the nominal trajec
tories zvp The asynchronous computation scheme
therefore enables running the primary process in closed
loop with system 1 at a high frequency The compu
tationally expensive secondary process then runs con
currently at a lower frequency and reduces conservative
ness by periodically providing new error dynamics pa
rameters and terminal set scalings Apart from limited
computational resources practical applications also im
pose memory constraints ie we can only store a lim
ited number of error dynamics parameters and terminal
set scalings which we handle by specific memory update
rules Figure 1 provides an overview of this scheme
51 Secondary Process Tube Optimization
The main objective of the secondary process is to com
pute error dynamics parameters ΦeΦνΣΞ and ter
minal set scaling α which define the state input and
disturbance tubes and the terminal set Therefore we
can choose the cost function of the secondary process ac
cording to any criterion the tubes and the terminal set
should fulfill eg a cost on Φepromotes more aggressive
tube controllers thus leading to smaller error bounds on
the state trajectories For more details on possible cost
functions and their effect on the tubes we refer to 28
Section 32 We formulate the secondary process simi
larly to 32 but with a modified objective ie
min
ΦeΦνΣΞα0LΦeΦνΣΞ 42a
st 32b 32l 42b
where L is the chosen cost function After termination
of 42 we use the optimized error system responses Φe
Φνto compute the tightened state and input constraints
fori 0 Nas
ZiX F iΦeViU F iΦν43
Additionally we use the optimized disturbance fil
terΣΞto compute the disturbance tubes for d
1 nDandi 0 N1 as
Qd
iσi1W F iΨd 44x Primary
Process 46
Memory M
Size MΦxeΦue
ΣΞ αSecondary
Process 42u
Fig 1 Visualization of the asynchronous computation
scheme for filterbased SLTMPC the secondary process com
putes error system responses ΦeΦν disturbance filter ΣΞ
and terminal set scaling α which are then passed to the pri
mary process and stored in memory M
where
FiΨdW i1M
j0Ψd
ijW
and Ψd
ijdefined in 18 For iN the disturbance
tubes are computed as
Qd
NσNW F NΨd 45
These sets are then passed to the primary pro
cess together with the error system responses the
disturbance filter and the terminal set scaling as
Mnew ZiViQd
i αΦeΦνΣΞ Since the mem
oryMMmm 0 M1 where we denote the
mthmemory entry as Mm is finite we need to define
a procedure that updates the memory whenever the
secondary process has computed a new Mnew thus ren
dering the memory timevarying In case the memory
is full the proposed update procedure in Algorithm 1
selects a memory slot Mmto store Mnewaccording to a
heuristic FMk For example FMk can choose the
slot with the least importance to the solution of the
primary process or assign a score to each memory slot
based on its historical importance and then choose the
slot with the lowest score
Algorithm 1 Update memory MSecondary Process
Input MnewMk1λk1
Output Mk
1procedure updateMemory Mnew