Lakhasly

SBGC-LSTM io gathcr dynamic data.En. Eurygasieri's location in the D-dimensional scarching spacc denoted wih the by the cxpresslon optinization, solution The fommula for Eurygaster >peed Each Eurygaster moves a (51,52, dhflcrent speed as they approach the cxlreme global value. Furthemmore.S, = (s,1, 5,2,.... C TEMPORAL HIERARCHICAL ARCHITECTURE Pollowing LSTM layers, SBGC-LSTM layers receive scnes where AT E of augmcnted node fcatures Rvd,. Three SBGC-LSTM layers are slacked in the pro- posed moxdcl to leam the temporal dynamics and spatial arrangement We creaie a lcmporal hicrarchical design of SBGC-LSTM with average pools in icmporal domains and inspired by spatial pooling o CNNS. Due lemporal hierarchical architecture, input temporal receplive ficlds of top SBGC-LSTM layers become short-term fasteners and are more sensitive to lcmporal dynamics. Morcover. they drastically reduce computing costs while cnhancing periomances [27).Particlc swarm optimization (PSO) and foraging icchniques are used in EOA optimizations, which are modelled afier curygasters: Based on the obseration that curygastcrs use thcir anicnnae to scan their sumoundings, this is truc.Thesc paramcters are initially set to high values, which decrease progressively: as a result, one trics t0 atlain a wide region helore rcducing to obiain a capacity that is reasonable ior Eurygasicr.The mark for the correctly dctecied position is qn. whereas the lefi detccicd position is denoicd by qk. These places include food flavor, which is represenied by the fitness function values /(qn) and /(qn), which were compulcd using the recommended method.The EOA aligorithm's location and speed updating procedure is as follows: atlats 820 where in Eq (14), (A/), is the cxpansion charactcristic of combination i at time t. The main point here is that the lincar portion and LSTM are distnbutcd among several charaeteristics.I) LEARNING SBGC-LSTM AE the cnd, the GF and of time stamp ane convert as resulis of and (owr); for CI phunses, where (ou), (ou ),.(ox)a.....Thcrefore, to remove the scale viriation between the two [eatures, an LSTM layer was applicd: EM =fomlconcara, Va) = fmlconcarn -fi-1) (14) the decaying weight coefficicnts.

SBGC-LSTM io gathcr dynamic data. Concalcnations of both
[catures as increasc the fcature information lo consider both
advantages. When fcature /A and frame diflerence fcature
Va are combined, the scale variations of the fcature vectors
are combincd howevcr, occur. Thcrefore, to remove the scale
viriation between the two [eatures, an LSTM layer was
applicd:
EM =fomlconcara, Va) = fmlconcarn -fi-1)
(14)
the decaying weight coefficicnts. l should be emphasized
d
tha .the pnobuibilivyd o only the
the latest
time stcp is utilizcd lo determine the iliness class.
2) EOA BASED TUNING PROCESS
The EOA was uscd io optimize the SBGC-LSTM sctings.
Particlc swarm optimization (PSO) and foraging icchniques
are used in EOA optimizations, which are modelled afier
curygasters: Based on the obseration that curygastcrs use
thcir anicnnae to scan their sumoundings, this is truc. The
strongest lood scenis are concentraled in the direction of the
anlennae; thus, Eurygasler moves in that dircction. A mcl
hcuristic optimization icchnique was developed by Marini
and Walczak I28)I. based on curygasier bchavior. The sicps
required io optimize SBGC-LSTM using the EOA are shown
in Fig 6. The dataset is divided into sevcral caiegories using
a kemcl-based clustering approach. Furthermore, the cluster
centcrs were uscd io initializc the fuzzy rules for SBGC-
LSTM. The SBGC-LSTM model was trained using an EOA
model. The location of Eurygasler in Q-dimensional space a
1 was delcrmincd using Eq (18).
4t=4 +y0(qn)-f(qm)sbes
(8)
In the above Eq (19), dir denotes the Eurypasier-secking
direction, which was mandomly sclccied. The lengths of
Eurygasicr scarch steps are depicied by sst, while disr
represcnt distances that antcnnac cncounter in Eq (20). Thesc
paramcters are initially set to high values, which decrease
progressively: as a result, one trics t0 atlain a wide region
helore rcducing to obiain a capacity that is reasonable ior
Eurygasicr. The mark for the correctly dctecied position is qn.
whereas the lefi detccicd position is denoicd by qk.
These places include food flavor, which is represenied
by the fitness function values /(qn) and /(qn), which were
compulcd using the recommended method. sf is the function
of the symbol. The update of speed and data collecied by
the antennas has an influence on the position of Eurygaster
Consider the Eurygasicr swarm atnbutes are updaicd using
E (21) 10 Eq (22) such that E = E..... En. Eurygasieri's
location in the D-dimensional scarching spacc
denoted
wih the
by the cxpresslon
optinization, solution The fommula for Eurygaster >peed
Each Eurygaster moves a
(51,52,
dhflcrent speed as they approach the cxlreme global value.
Furthemmore.S, = (s,1, 5,2,.... sD)" isused lo designale the
cxtremes of cach Euryghasier. E = (cxl. ere,....exp)"
descnibes global extremes. The EOA aligorithm's location and
speed updating procedure is as follows:
atlats 820
where in Eq (14), (A/), is the cxpansion charactcristic of
combination i at time t. The main point here is that the
lincar portion and LSTM are distnbutcd among several
charaeteristics.
C TEMPORAL HIERARCHICAL ARCHITECTURE
Pollowing LSTM layers, SBGC-LSTM layers receive scnes
where AT E
of augmcnted node fcatures
Rvd,. Three SBGC-LSTM layers are slacked in the pro-
posed moxdcl to leam the temporal dynamics and spatial
arrangement We creaie a lcmporal hicrarchical design of
SBGC-LSTM with average pools in icmporal domains and
inspired by spatial pooling o CNNS. Due
lemporal
hierarchical architecture, input temporal receplive ficlds
of top SBGC-LSTM layers become short-term fasteners
and are more sensitive to lcmporal dynamics. Morcover.
they drastically reduce computing costs while cnhancing
periomances [27).
I) LEARNING SBGC-LSTM
AE the cnd, the GF and
of time stamp ane convert
as resulis of and (owr); for CI phunses, where (ou),
(ou ),.(ox)a..... (onf ).c ).and finding probabilities being
phases are oblained in below Eq (5).
=1,2..d)
(5)
Assuming that cach atrnbule slep in the top SBGC-LSTM
is concenlcd during training and has short-term dynamics,
we supervisc the modcl with the following loss funetion.
C
Loss
(
(16)
where in Eq (16), the ground-truth label is y = (yi.....ycl).
T indicates the / SBGC-LSTM layer time-stcp count.
The third icmm givescvery joint cqual consideration. The final
clause restricts the number of interesied nodes. and represent
VOLUMAE 12, 2024
26689