Hybrid BSO+BMO: An Optimal Routing of Adaptive Deep Learning Approach with Attention Mechanism for Link Failure Detection in Elastic Optical Network

Abstract

Because of the enhancement in the data center services, the “Elastic Optical Network (EON)” is a very successive framework to interlink the information centers. The EON can elastically provide a spectrum tailored for multiple needs of bandwidths. In the link failure case, confirming the high stage “Quality of Service (QoS)” for candidate requests after the fault leads to an experiment focus. With the help of the modern digital signal processing approaches and developments in the integrated circuits and the coherent receivers in EON is able to estimate the link failures in the present time. The high-speed network survivability is highly important. When the sizes of the network get enhanced, the likelihood of the node and link impairment is also enhanced. Therefore, to predict the link impairment in EON, an adaptive technique is necessary. To accomplish this objective, a novel methodology is proposed using hybrid heuristic improvement. In the first stage, the required data is gathered and fed into the link failure detection model. The novel method is named an Atrous Spatial Pyramid Pooling – 1 Dimensional Convolution Neural Network with Attention mechanism (ASPP-1DCNN-AM), in which some of the hyper-parameters are tuned by proposing the hybrid algorithm as Iteration-aided Position of Beetle and Barnacles Mating (IPBBM). After forecasting the failure link, the model is in need of finding the optimal routing for better communication. Here, the optimal path is identified by using the IPBBM algorithm. Finally, the validation is done using divergent measurements and in contrast with traditional models. Hence, the designed system demonstrates that it achieves the higher detection results to make the data transmission effectively.

Full text

Hybrid BSO+BMO: An Optimal Routing of Adaptive
Deep Learning Approach with Attention Mechanism
for Link Failure Detection in Elastic Optical Network
Sharma Sahana (  sahanasharma161@gmail.com )
KS Institute of Technology
K.V.S.S.S.S Sairam
NMAM Institute of Technology
Research Article
Keywords: Elastic Optical Network, Link Failure Detection, Optimal Routing, Atrous Spatial Pyramid
Pooling – 1 Dimensional Convolution Neural Network With Attention Mechanism, Iteration-Aided Position
Of Beetle And Barnacles Mating
Posted Date: July 3rd, 2023
DOI: https://doi.org/10.21203/rs.3.rs-3114667/v1
License:   This work is licensed under a Creative Commons Attribution 4.0 International License. 
Read Full License
Additional Declarations: No competing interests reported.

Hybrid BSO+BMO: An Optimal Routing of Adaptive Deep Learning
Approach with Attention Mechanism for Link Failure Detection in Elastic
Optical Network
1Sharma Sahana, 2Dr. K.V.S.S.S.S Sairam
1Assistant Professor
Artificial Intelligence and Machine Learning
KS Institute of Technology, Visvesvaraya Technological University
TU- Belagavi Bangalore-560109, India
sahanasharma161@gmail.com
2Professor & Head
Electronics and Communication Department
NMAM Institute of Technology, Visvesvaraya Technological University
Nitte, Karkala, Udupi, Karnataka- 574110, India
Abstract- Because of the enhancement in the data center services, the “Elastic Optical
Network (EON)” is a very successive framework to interlink the information centers. The
EON can elastically provide a spectrum tailored for multiple needs of bandwidths. In the link
failure case, confirming the high stage “Quality of Service (QoS)” for candidate requests after
the fault leads to an experiment focus. With the help of the modern digital signal processing
approaches and developments in the integrated circuits and the coherent receivers in EON is
able to estimate the link failures in the present time. The high-speed network survivability is
highly important. When the sizes of the network get enhanced, the likelihood of the node and
link impairment is also enhanced. Therefore, to predict the link impairment in EON, an
adaptive technique is necessary. To accomplish this objective, a novel methodology is

proposed using hybrid heuristic improvement. In the first stage, the required data is gathered
and fed into the link failure detection model. The novel method is named an Atrous Spatial
Pyramid Pooling – 1 Dimensional Convolution Neural Network with Attention mechanism
(ASPP-1DCNN-AM), in which some of the hyper-parameters are tuned by proposing the
hybrid algorithm as Iteration-aided Position of Beetle and Barnacles Mating (IPBBM). After
forecasting the failure link, the model is in need of finding the optimal routing for better
communication. Here, the optimal path is identified by using the IPBBM algorithm. Finally,
the validation is done using divergent measurements and in contrast with traditional models.
Hence, the designed system demonstrates that it achieves the higher detection results to make
the data transmission effectively.
Keywords- Elastic Optical Network; Link Failure Detection; Optimal Routing; Atrous
Spatial Pyramid Pooling – 1 Dimensional Convolution Neural Network With Attention
Mechanism; Iteration-Aided Position Of Beetle And Barnacles Mating
1. Introduction
The EON enhances on the restricted spectrum utilization in the conventional "Wavelength-
Division Multiplexing (WDM)"-aided optical structures by employing the flexible allocation
of spectrum to transfer better transport of hundred Gbit/s features and above [1]. EONs can
complete the throughput of the single fiber link to correspondingly 10 to 100 Tb/s. The
optical structures have attained huge importance because of their capability to handle very
huge information rates utilizing the "dense wavelength division multiplexing" methodology
[2]. With certain huge information rates, a small severe disturbance in the approach of the
model can lead to a high amount of information loss. Generally monitored several
disturbances are created by the human errors, excessive bit errors, equipment failure, and the

fiber cuts [3]. It is considered that these errors be specially recognized and solved in the
physical layer prior to they are identified at the upper layers. Thus, it is complex for the
optical networks to utilize effective and fast approaches for recognizing and identifying the
failures in networks [4]. Several failures like "cross-connect port intrusion and blocking" can
trouble a separate or a particular subset of wavelengths present in the link [5]. Other faults,
consisting of "Bit Error Rates (BERs)" and fiber cuts can trouble entire wavelengths that
transmit via a fiber duct.
In an optical network methodology, a failure of the link is localized and recognized easily
depending on the on-off criteria of the signals of optical supervisory [6]. The hardware
expense of the supervisory channels and the monitors is essential for attaining speedy failure
in link localization at the optical layer and is to be reduced in the model [7]. In the meantime,
decreasing the needed amount of monitor’s are results in minimal fault handling works.
Because of the inefficiency in the “All-Optical Networks (AONs)” electrical terminations, the
fault identification approaches developed for conventional optimal models may not be
exchanged for AONs. Subsequently, the detection of a fault in the AONs may be developed
at distinct protocol layers. In the network layer, high-routing protocols have the ability of
built-in error identification [8]. To minimize the period of identification, the cross-layer
models are also utilized [9]. But, certain approaches can provide only an identification period
in seconds that is very longer than the general necessity of 50 ms for the recovery in the
optimal networks [10].
Moreover, the occurrences in the EON models like the disruption of transmissions for
several hundreds of candidates, and optical fiber cuts create a data and revenue loss [11].
Various experiments have concentrated on enhancing the service presence of EONs, WDM,
or huge layers over failures [12]. The EON's is not having the limited wavelength channels
which are intrinsic in the conventional WDM structures [13]. Importantly, the EON suggests

multiple grids that are distinct from the limited grid of conventional WDM models. The slot
dimensions have referred to by the "ITU-T G.709" that is the generally considered 12.5 GHz
[14]. Several faults like "Optical Cross Connect (OXC) port blocking" and "transmitter lease
failure" trouble the separate or particular wavelengths. This type of error may be managed by
the in-place error handling or control sectors [15]. Other errors can trouble entire fiber
wavelengths like “Optical Amplifier (OA)" saturation and fiber cuts and so on. Without the
generality loss, it is named a link failure. To offer a period of critical recovery, the speedy
link failure identification is necessary [16].
The important role of the recommended link failure detection system in EON is
summarized here.
• To implement a link failure detection system in EON that effectively recognizes
the failed link and helps to transmit the data optimally with the utilization of the
optimal routing approach.
• To detect the failure link successfully employed the ASPP-1DCNN-AM system
that integrates the 1DCNN and ASPP with an attention approach. Here, the hyper-
parameters are tuned by the improved IPBBM.
• To improve the IPBBM that mimics the concepts of traditional “Beetle Swarm
Optimization (BSO)” and “Barnacles Mating Optimizer (BMO)” and supports
parameter optimization and optimal path selection.
• To choose the optimal path effectively for the network adopted the developed
IPBBM system that helps to transmit the data in a safe and secure manner with a
minimal time.
• To calculate the system functionality adopted divergent traditional classifiers and
optimization algorithms with the utilization of several performance metrics.

The research work on link failure detection systems in EON includes upcoming parts. Part
II contains the existing tasks of the recommended link failure detection model. Part III
illustrates the optimal link failure detection and routing for the EON. Part IV demonstrates
the ASPP-1DCNN-AM for identifying the link failure. Part V describes the implementation
of IPBBM for parameter optimization. Part VI elaborates on the results of the developed link
failure detection system in EON with various generated networks. Part VII summarizes the
implemented work of link failure detection in EON.
2. Existing Works
2.1 Related Works
In 2011, Wu et al. [17] have recommended a modern method called "Minimum-Length M-
Cycles (M2-Cycles)" to generate a cycle cover that included a collection of "M2-Cycle”.
Scholars confirmed that the designed system attained a similar degree of localization as the
method of spanning tree depended. But that needed a minimal number of assisting assets
even though the spanning tree was enlarged. The experimental outcomes assured that the
theoretical evaluation and viewed that the assisting assets needed by the developed model
were minimized.
In 2009, Ahuja et al. [18] considered the issue of error localization in the entire AONs.
Researchers presented the idea of “Monitoring Paths (MPs)” and “Monitoring Cycles (MCs)”
for the separate recognition of the failures in the separate link. For a system with a separate
observing region, experts confirmed that the connectivity of three-edge was important and
enough criteria for developing the MCs. Moreover, experts formulated the issue of generating
the MCs as an "Integer Linear program (ILP)". Also, experts implemented a heuristic model
for generating the MCs in the existence of multiple observing regions. From the simulation
findings, experts demonstrated the efficiency of the recommended observing methodologies.

In 2023, Assis et al. [19] have explored a Mixed ILP (MILP) validation to locate the
unequally divided connection's rate in the transmission between the various disjoint ways to
offer a minimized amount of network slots and better protection. The final answers were
evaluated over multiple realistic networks and displayed that the developed estimation
surprisingly reduced the spectral usability contrasted to the existing protection approaches
and decreased the needed squeezing count.
In 2016, Mohan et al. [20] have presented a model to focus on the AON's survivability in
the failure of separate links. The model reduced the recovery period of the failure in a
separate link. The "Ant Colony Optimization (ACO)" methodology, accompanied by the
neighboring minimized cycle was employed to evaluate the better path for data
retransmission.
In 2017, Vela et al. [21] have experimented with some of the reasons for failures that
troubled the standard of the optimal links and recommended two distinct models. The initial
model concentrated on identifying the BER modifications in the optical links called bando.
The second one is concentrated on detecting the highest possible pattern of failure called
lucida. The research was performed for the two distinct models to achieve measures for the
attained power and BER and utilized to create the synthetic information utilized in the next
experiments.
In 2020, Shu et al. [22] have deployed the digital spectra feasibility in utilizing the "Soft
Failure Identification (SFI)" and "Soft-Failure Detection (SFD)". Scholars recommended a
digital spectrum-aided SFI and SFD architecture. The dual-phase SFD model was utilized to
minimize the processing and observing of the optical node overheads. The numerical
outcomes were provided to evaluate the features of the digital spectrum and attribute
distributions of four general soft faults. In the end, the experiment findings estimated the
identification and detection functionality of the designed model.

In 2022, Sun et al. [23] have offered a detection model and SFD according to the digital
residual spectrum, which resisted the "Support Vector Machine (SVM) and Auto-Encoder
(AE)". The features of the model were low cost, easy training, and high generalization.
Utilizing the system learned for a particular setup, experts illustrated a region in the detection
accuracy and curve.
In 2022, Bao et al. [24] have suggested an approach called "Link-Oriented Resource
Balancing (LoRB)" in the EONs that allocated the life path request spectrum based on the
minimal path. The solutions represented that the developed LoRB model could decrease the
blocking likelihood and enhance the usage of resources contrasted with the existing EONs.
2.2 Research Gaps and Challenges
The EON is a successive answer to face the future generation's larger bandwidth requirement.
When the size of the optimal network gets enhanced, the likelihood of dual and single-link
failure is also enhanced. In the EON, the failure of the link creates a large loss of data and
high disruption to the candidates. The link failure in the network creates a poor QoS for the
customers and also creates revenue loss for the operator of the network. This also creates high
congestion in the network. So, the detection of link failure in the EON is necessary. Several
research works have been developed to investigate these issues, but still they need
improvements. Table 1 elucidates several challenges and features of the conventional link
failure detection model in EON. The spanning tree-based approach [17] provides better
performance even if the tree gets enlarged. It can extend the nodes of the network. However,
it is a complex structured model. It is not focused on the system's efficacy. The “Monitoring
Cycle Problem (MCP)” algorithm [18] locates the set of monitoring paths. It eliminates the
unnecessary edges of the network. But, it is very inefficient. It requires more computational
resources. Multipath routing [19] prevents the data traffic between two nodes. It allows

independent similar traffic squeezing. Yet, it has limited end-end capacity. It enhances the
complexities of the model. ACO [20] is utilized to estimate the alternative path of the
network. It helps to solve the combinatorial optimization issues. However, it has dimensional
issues. It is not applicable to the large datasets. BANDO [21] detects the BER changes in the
optimal connections. It enhances the detection of gradation. But, it produces the inaccurate
data. It has underfitting issues. Gaussian distribution-based anomaly detection algorithm [22]
solves the anomaly detection issues. It has low complexity. But, it degrades the functionality
when the iteration count is high. It is not applicable to the real-time applications. AE [23]
minimizes the noise of the given data. It eliminates the complexity of the datasets. But, it is
not considering the necessary data. It provides unrealistic solutions. LoRB scheme [24]
enhances the resource utilization. It decreases the blocking probability. But, it misclassifies
the input errors. It works only for a limited amount of inputs. To handle these hurdles, it
implemented a new model for detecting the link failure in the EON utilizing hybrid
optimization models.
Table 1. Features and challenges of existing link failure detection model in EON
Author
[citation]
Methodology
Features
Challenges
Wu et al.
[17]
Spanning tree-based
approach
• It provides better
performance even
the tree gets
enlarged.
• It can extend the
nodes of the
network.
• It is a complex
structured model.
• It is not focused on
the system's efficacy.
Ahuja et
al. [18]
MCP algorithm
• It locates the set of
monitoring paths.
• It eliminates the
unnecessary edges of
the network.
• It is very inefficient.
• It requires more
computational
resources.
Assis et al.
[19]
Multipath routing
• It prevents the data
traffic between two
nodes.
• It allows independent
• It has limited end-end
capacity.
• It enhances the
complexities of the

similar traffic
squeezing.
model.
Mohan et
al. [20]
ACO
• It is utilized to
estimate the
alternative path of
the network.
• It helps to solve the
combinatorial
optimization issues.
• It has dimensional
issues.
• It is not applicable to
the large datasets.
Vela et al.
[21]
BANDO
• It detects the BER
changes in the
optimal connections.
• It enhances the
detection of
gradation.
• It produces the
inaccurate data.
• It has underfitting
issues.
Shu et al.
[22]
Gaussian distribution-
based anomaly
detection algorithm
• It solves the anomaly
detection issues.
• It has low
complexity.
• It degrades the
functionality when
the iteration count is
high.
• It is not applicable to
the real-time
applications.
Sun et al.
[23]
AE
• It minimizes the
noise of the given
data.
• It eliminates the
complexity of the
datasets.
• It is not considering
the necessary data.
• It provides unrealistic
solutions.
Bao et al.
[24]
LoRB
• It enhances the
resource utilization.
• It decreases the
blocking probability.
• It misclassifies the
input errors.
• It works only for a
limited amount of
inputs.
3. Illustration of Optimal Link Failure Detection and Routing Mechanism for
Elastic Optical Network: Attention-based Deep Learning Model
3.1 Data Generation
The data generation approach is performed after the approach of network creation. Before
initiating this approach, the factors such as traffic volume, long link, and total length, number
of links, BER, modulation, nodes position, and lambda are taken into the consideration. The
network will be overabundance if the EON’s density is large. The EON includes an enormous

amount of information. By utilizing this data generation approach, it can be managed in an
easier manner. This approach is mainly employed to find out whether the connection contains
a failed link or not. The data generation factors are highly useful for identifying the failures
of links with maximum accuracy. The generated data is indicated as
EON
d
S
.
3.2 Link Failure and Routing Issues in EON
Issues in link failure:
The capacity and size of the EON are enlarging rapidly, so any kind of fault in the
EON leads a data and service loss. So, the survivability of the communication
network is very complex.
The tolerance of dual and single-link failure is helpful, yet the network still
remains with the multi-link failure risks.
The multi-link failure is very general in real-time networks. Because of the long
period of consumption to repair a physical link cut leads to a multi-link failure.
The reliability of every metric in the optical network minimizes as its time of
utility gets enhanced. This enhances the likelihood of multiple-link failure.
The earthquakes, floods, and fires will also create a high amount of nodes and
links degraded and result in link failure.
The protection task should handle the link failures, most importantly for those that
have high robustness and security.
Failure recovery is an important issue. The hard failure initiates the formation of
soft failure. After that, they will create a large problem for the network. So, the
failure of the link is handled by the routing methodology.
Issues in routing:

✓ The existing routing methodologies do not offer the standard communication, and
it takes much time for transmission.
✓ Also, the existing routing tasks consumed more cost and it does not offer an
effective path.
✓ It also leads of outages in a network and troubles to find a alterative path for the
recovery.
3.3 Proposed EON Model of Detection and Routing
The optical networking methodologies are very essential to the global internet approaches
and their capability to help the reliable and critical transmission services. However, the
optical networks executing in a limited wavelength grid importantly allocate an entire
wavelength to the need of network traffic that does not complete its overall capacity. This
ineffective usage of spectral assets may have become a more complex problem with the
development of large information rates. The EON has the ability to solve the coarse and
limited granularity of conventional optical methodologies and is assumed to help the flexible
information rates and demands of variable bandwidth effectively. But, it also brings modern
problems like fragmentation of bandwidth to robust, management and control of the effective
network, and link failures. The basic flow of the implemented link failure detection model in
EON is presented in Fig.1.
Figure 1. The basic flow of the improved link failure detection model in EON
Generated
data
Link failure
detection
ASPP-
1DCNN-AM
Parameters
are optimized
by IPBBM
Optimal
routing
approach
Optimal route
selection by IPBBM
Better
transmission

The detection of the link impairment model in EON is constructed to forecast the link
failure. To achieve this goal, a technique is recommended utilizing hybrid heuristic
development. In the initial phase, the necessary data is collected and subjected to the link
failure identification system. The approach is named ASPP-1DCNN-AM in that certain
hyper-parameters are optimized by the hybrid model called IPBBM for identifying the failure
link. After detecting the link failure, the model has the necessity to locate the optimal routing
for good transmission. In this, the optimal path is detected by utilizing the developed IPBBM.
In the end, the estimation is performed on the divergent metrics and contrasted with the
existing systems. Thus, the recommended model illustrates that it accomplishes the better
detection solutions to create the data communication efficiently.
4. ASPP-1DCNN with Attention Mechanism for Detecting the Link Failure in
EON and Parameter Optimization
4.1 Basic 1DCNN
The framework of 1DCNN [25] is utilized to detect the link failures in the EON. From the
data generation approach, created data
EON
d
S
is fed into this technique. The basic 1DCNN
includes of some layers called pooling, dropout, and convolution layers. To generate the
1DCNN structure, the every layer’s neuron counts, sub-sampling factor, and the filter size is
utilized. The filtering approach is performed with the aid of the layer called convolution. The
simultaneous usability of the filtering operation results from the attribute map and it contains
of factors from the information point. The convolution approach is processed by multiplying
the measures of input and kernel factor. The factors such as filter amounts and epochs are
optimally determined to achieve the correct data feature. The resultant measure is employed a
few times to process the approach and it provides various measures. These measures are
creating the feature map. The measures from the feature map are sent to the “Rectified Linear

Unit (ReLU)” function. Because of this function, good functionality is achieved and it also
handles the issues of gradient. It is formulated in Eq. (1).
( ) ( )
EON
d
EON
dSST ,0max=
(1)
The variable
EON
d
S
denotes the activation function’s input and
( )
EON
d
ST
refers to the output of
the positive activation function. Another layer of the 1DCNN is called the pooling layer. This
layer creates the mapped pool attributes by performing the attribute maps. The attributes such
as stride, kind of pooling, and the pooling filter size are taken into consideration to select the
mapped fool attributes. The mean and max pooling are employed to estimate the mean and
maximum measures of the entire patch accordingly. The entire concert of this network is
minimized by utilizing the deep learning methodologies. Hence, the dropout layers are
employed to attain better functionality rates. Here, the input measure is initially set as zero
for the entire steps. Thus, it avoids the overfitting of the system and it is expressed in Eq. (2).
su
SEON
d−
=1
1
(2)
The drop-out layer is chosen by utilizing scaling. However, the dropout layer enhances the
weight of the network. The 1DCNN framework is given in Fig.2.
Figure 2. The framework of the 1DCNN for the link failure detection model in EON
Soft
max
1D Input
Con.1
Hidden
layer
CNN Architecture
Dense
layer
Con.4
Con. 3
Con. 2
Fully Connected Layer

4.2 ASPP-1DCNN with AM for Detection
The 1DCNN structure supports minimizing the unnecessary data and provides suitable data
for the link failure detection. However, it has high computational complexity and increases
the network weight. To solve these problems in the improved link failure detection system in
EON it is integrated with the ASPP architecture. The ASPP [26] is a parallel architecture of
certain branches that perform to the similar attribute map and join the results at the end stage
and it was presented in DeepLab's second version architecture. The ASPP utilizes the atrous
convolution in the entire branch. The ASPP varies from the standard convolution with its rate
parameter that joins the specific amount of zeros among the attributes in the convolution
filter. This approach equals the operation of downsampling, upsampling, and convolution yet
has more good functionality without enhancing the parameter amounts that handle the
network efficacy. The ASPP architecture has two kinds. The real one in the DeepLab second
version contains four branches of atrous convolution with values of 6, 12, 18, and 24.
However, the convolution filter with value 24 is very similar to the input attribute map
dimensions, only the middle of it performs the process. In the DeepLab third version was
exchanged by the 1x 1 convolution. Furthermore, a pooling branch is also joined to utilize the
overall context data. The ASPP is recommended to fuse the high-level data created by the
atrous convolution with distinct dilation values that enhances the attribute receptive fields to
a particular amount. However, the approach is infeasible. So, the ASPP-1DCNN network is
integrated with AM to produce the better solutions. The AM permits the system to
concentrate a particular section of input by allocating a distinct weight to divers sections of
the input. It helps to utilize the more related way of input sequences in a better way. The
formulation of the AM is expressed in Eq. (3).

( )








=
k
t
v
kequ
softvakequA*
max,,
(3)
The variables
quandkeva,
denote the value, key, and query matrices accordingly. The
ASPP-1DCNN model is processed with the real data
EON
d
S
. The developed ASPP-1DCNN-
AM system helps to achieve the efficient solutions and reduces the processing time. It also
minimizes the computational burdens. The architectural view of the ASPP-1DCNN-AM is
illustrated in Fig.3.
Figure 3. The architectural view of the ASPP-1DCNN-AM model for the developed link
failure detection system in EON
4.3 Parameter Optimization in Detection Model
When processing the ASPP-1DCNN-AM system for the link failure detection, the system
may have trouble producing an error-free solution. This issue will be resolved after involving
certain parameters. The parameters such as epoch, hidden neuron counts, and steps per
epochs are involved in this operation. These parameters are tuned with the aid of improved
Input data
Soft
max
1D Input
Hidden
layer
CNN Architecture
Dense
layer
Fully Connected Layer
1x1 conv
3x3 conv
rate 6
3x3 conv
rate 12
3x3 conv
rate 18
Image
pooling
Attention Mechanism
1x1
conv
Concat
ASPP-AM
ASPP-1DCNN-AM
Link failure detected
outcome
Hidden neuron,
epoch, steps per
epoch are
optimized by
IPBBM
Maximization of accuracy, MCC, and
minimization of FPR

IPBBM. This algorithm enhances the functionality of the developed model and improves the
detection ability. With the support of this approach, the developed link failure detection
model achieves the classified results for the generated data.
5. Explaining the Iteration-aided Position of Beetle and Barnacles Mating for
Optimal Routing and Objective Function
5.1 Tuning by IPBBM
The existing BSO and BMO are combined to produce the designed IPBBM. The classical
BSO is enhanced from the idea of conventional “Particle Swarm Optimization (PSO)” and
the “Beetle Antenna Search (BAS)”. Similarly, the BMO is mimicked from the features of
the microorganism named a barnacle. The BSO can overcome the issues of poor stability and
convergence. The BMO provides the standard solutions concerning the exploration,
convergence features, and exploitation. However, the BSO lacks to offer effective solutions
also the BMO slows down the process when the iteration count enhances. Therefore, the
IPBBM is presented to solve the classical model problems. In this approach, the maximum
iteration is divisible by two and that value is greater than the present iteration means the BSO
will be performed. Or else, the BMO will be performed. It is mathematically presented in Eq.
(4).
2
max
U
u
(4)
Here, the parameters
max
Uandu
denote the present and maximum iterations
correspondingly. The functionality of the conventional BSO and BMO are presented below.
BSO: The traditional BSO [27] is the integration of the PSO and BAS. The separate
candidates in the BSO contrasted the fitness measure of the right and left side measures while

every execution and contrasted the better measures of the two. The beetle swarm upgraded
expressions are presented from Eq. (5) to Eq. (7).
( ) ( )( )
musu
u
jyyfsigncwc −−= ..

(5)
( ) ( )
1321
11 ...... wcrddyQhrddyQcrddww i
j
i
j
i
j
i
j
i
j
i
j+−+−+= ++
(6)
11 ++ += i
j
i
j
i
jwyy
(7)
Here, the variable
1+i
j
w
indicates the
th
j
particle speed after the
th
i
execution, and the term
1
+
i
j
y
points to the
th
i
execution place. Then, the attribute
j
wc
refers to the rate of updating
created by the BSO model. The symbolic function and the learning factors are denoted as
( )
321 ,, candccsign
correspondingly. Furthermore,
rd
denotes the arbitrary measure and
Q
is
the stock price indication. The size of the step is pointed as

. The stages of the BSO are
described below.
At first, the algorithm attributes, learning factors
321,candcc
, the weight of inertia
X
, and
the distance
0
e
are initialized among the two antennae for every beetle.
Then, randomize the velocity
w
and place
y
and estimate the fitness measure of every
place. The present global optimum and the separate optimal answer are termed as
bestbest QandH
appropriately.
Next, randomize the direction of the beetle heads. Estimate the fitness of right and left
beetle for the population and it is expressed in Eq. (5).
Then, the present upgrade criteria for the every beetle are estimated and it is given in Eq.
(6).
Then, the overall region-updated criteria conditions are expressed in Eq. (7).
BMO: The existing BMO [28] is developed from the unique features of barnacles which is
a micro-organism. The functionality of this model is presented as follows. The member
solution
Y
of the BMO is formulated in Eq. (8).











=
M
mm
M
yy
yy
Y



1
1
1
1
(8)
The control measure and the population number are termed as
nandM
accordingly. The
“lower
Lo
and upper
Up
” limits of the control measures are estimated in Eq. (9) and Eq. (10)
appropriately.
 
j
loloLo ,...,
1
=
(9)
 
j
upupUp ,...,
1
=
(10)
Based on the below assumptions the selection approach will occur.
o The choosing operation is carried out arbitrarily but it will be stopped to the
barnacle penis length
ql
.
o Every barnacle may grant its sperm and also can attain the sperm from a distinct
barnacle.
o If at a particular stage, the selection approach chooses the same barnacle means,
that will not be considered for the process.
o If the choosing approach at the specific execution is higher than the provided
ql
,
then the sperm cast approach will occur.
From Eq. (11) and Eq. (12) can learn that the selection approach is happened randomly
and that attains the initial assumption.
)(_ mpermrdbbarnacle =
(11)
)(_ mpermrdgbarnacle =
(12)
The terms
gbarnacleandbbarnacle __
refer to the mated parents.
The reproduction of the barnacle is mathematically given in Eq. (13).
newM
gbarnacle
newM
bbarnacl e
newM
jqyqyy_
_
_
_
_+=
(13)

Here, the factor
q
denotes the generally distributed arbitrary integer among 0 and 1. The
dad and mum of the barnacle are given as
newM
gbarnacle
newM
bbarnacl e yandy _
_
_
_
respectively and they are
chosen from Eq. (11) and Eq. (12).
The approach of sperm cast is formulated in Eq. (14).
m
gba rnacle
newM
jyrdy_
_=
(14)
Here,
rd
is the arbitrary integer which lies in [0, 1]. The functional chart and the pseudo-
code are given in Fig. 4 and Algorithm 1 accordingly.
Algorithm 1:IPBBM
Consider the overall populations and execution counts
Measure the objective function
For
max
1Utou=
For
ppn
mtoj1=
2
max
U
uif 
Execute the BSO
Estimate the left and right beetle’s
fitness by applying Eq. (5).
Update the position of the beetle
utilizing Eq. (7).
Else
Perform the BMO
Do the reproduction approach utilizing
Eq. (13).
Process the sperm cast operation
employing Eq. (14).
End if
End
End
Repeat the executions.
Accomplish the optimal solutions

Figure 4. The flow diagram of enhanced IPBBM algorithm
5.2 Optimal Routing and its Objective Function
After detecting the failure of the link effectively, the system should validate the better route
to transmit the data. In order to choose the route optimally, the system takes the help from the
improved IPBBM. Here the generated data
EON
d
S
is subjected as an input. With the assistance
of certain attributes, the optimal routing approach is conducted. To avoid the overfitting
issues in the approach the attributes are optimized by the improved IPBBM. The objective
function of this approach is presented in Eq. (15).
 






++
++++=
−−− DyDtFPR
TPDRMCCAob
DCNNAS PPDCNNASPPDCNNASPz seephnsp
1
maxarg
111 ,,,
(15)
Here, the hidden neuron count is pointed as
DCNNASP
hn 1−
and an epoch count in ASPP-
1DCNN is termed as
DCNNASPP
ep 1−
. Then, the shortest path is referred as
z
sp
and a step per
epoch in ASPP-1DCNN is denoted as
DCNNASPP
se 1−
. The hidden neuron count is ASPP-1DCNN
lies from 5 to 255. The epoch count in ASPP-1DCNN is varied from 5 to 50. The shortest
Best position updating by BMO
Start
Population Initialization
Fitness value evaluation
Return the optimal values
End
Yes
Best position updating by BSO
No
Yes
No
2
max
U
uif 
max
Uu

path altered from 1 to a number of nodes and the steps per epoch in ASPP-1DCNN limits
from 50 to 250. Moreover, the terms
TandPDRMCCA ,,
denote the accuracy, “Matthews’s
Correlation Coefficient (MCC)”, “Packet Delivery Ratio (PDR)” and throughput
correspondingly. Further, the factors
DyandDtFPR,
pointed to the “False Positive Rate
(FPR)”, distance, and delay accordingly. The solution encoding diagram of the developed
system is given in Fig.5.
Figure 5. The solution encoding diagram of the improved link failure detection model in
EON and optimal routing
5.3 Description of Constraint Used
The constraints which are helped to attain the optimal routing are explained here.
Accuracy:” The relation among the real and collected data”.
SRQPONML
ONML
A+++
+
=
(16)
MCC: “The changes among the real and detected data are referred to as MCC”.
))()()((SRMLQPMLONSRBASR
ONQPONSR
MCC ++++
−
=
(17)
FPR:” It is described as the factor that is recognized by the error”.
MLQP
QP
FPR +
=
(18)
IPBBM
Maximization of
Accuracy, MCC
and
Minimization of
FPR
Maximization of
PDR, throughput
and Minimization
of Distance,
Delay
DCNNASPP
hn 1−
DCNNASPP
ep 1−
DCNNASPP
se 1−
z
sp
❖
DCNNASPP
hn 1−
-Hidden neuron
count in ASPP-1DCNN
❖
DCNNASPP
ep 1−
-epoch count in
ASPP-1DCNN
❖
DCNNASPP
se 1−
-step per epoch in
ASPP-1DCNN
❖
z
sp
-Shortest path

In the above expressions, the “false negative and true positive” is termed as
MLandSR
accordingly. Further, the “negative and false positive” is pointed as
QPandON
correspondingly.
PDR: It is referred to as the “ratio of packet numbers that are delivered and the packet
numbers that are sent from source to destination”.
j
j
S
R
PDR =
(19)
Here, the term
j
R
denotes the received data and
j
S
indicate the transmitted data.
Throughput: “It is the ratio between the time and inventory” and it is expressed in Eq.
(20).
ti
in
T=
(20)
The attributes
inandti
point to the time and inventory accordingly.
Distance: It is the estimation of “how far apart the destination is from the source”.
Delay: “It is the measurement of how much time takes to process the particular data
packet in the data transmission”.
6. Results and Discussions
6.1 Simulation setup
The designed link failure detection system in EON was processed in Python and the
surprising outcomes were attained. The highest count of the recommended system was 50 and
the population was 10. Moreover, the length of the chromosome was 10. The improved
system was contrasted with certain existing optimization models such as “Mexican Axolotl
Optimization (MAO)-ASPP-1DCNN-AM [29], Tuna Swarm Optimization (TSO)-ASPP-
1DCNN-AM [30], BSO -ASPP-1DCNN-AM [27], and BMO-ASPP-1DCNN-AM [28]

correspondingly to validate the performance. Also, several classifiers such as “Long Short
Term Memory (LSTM) [31], Recurrent Neural Network (RNN) [32], Resnet [33] [34], and
ASPP-1DCNN [26] [25] were utilized to evaluate the system functionality.
Moreover, the simulation approach was performed with the aid of generated networks
such as “USNET, NSFNET, COST239, and ARPANET”. The link impairments in the model
are detected utilizing the mentioned networks in minimal path transmission. The summary of
these networks is described as follows.
❖ ARPANET: It is mostly employed to explain the data over very long distances and
also it is utilized to validate the network.
❖ COST239: It includes various servers and also includes enormous network links.
❖ NSFNET: It is employed for the transmission of the shortest path and it contains 21
connections and 14 nodes.
❖ USNET: This network is utilized for the path transmission and it has 24 nodes and
43 connections. It is very trustworthy.
6.2 Evaluation metrics
The functionality metrics for the link failure detection system in EON are explained here.
Precision: “It is termed as the factor of fault detected and the better answers”.
ONML
ML
P+
=
(21)
F1-score: “It is referred to as the recall and precision value rates”.
QPON
QPON
SF +

= 21
(22)
Sensitivity:” It mentions the changes of less absolute number that may be recognized as a
variable”.
ONML
ML
Sen +
=
(23)

FNR:” It calculates the faults that are detected in the pictures”.
MLSR
SR
FNR +
=
(24)
FDR:” It is described as the estimation of describing the FP, both TP and FP rates”.
QPML
QP
FDR +
=
(25)
Specificity:” It is termed as the negative ratio probability estimation”.
SRQP
QP
spec +
=
(26)
Recall:” It is indicated as the variable that calculates the optimal positive factor counts
over total positive factor”.
SRON
ON
+
=Re
(27)
6.3 Analysis of the confusion matrix for the implemented link failure detection model in
EON
Fig.6 displays the estimation of the confusion matrix for the improved link failure
detection model in EON. This estimation was conducted with the four networks such as
“USNET, NSFNET, ARPANET, and COST239” accordingly. With the aid of actual and
predicted measures, the networks are varied. From Fig.7 (c) and (d), it is confirmed that the
ARPANET and COST239 networks attained better accuracy i.e., 97.24% for the improved
link impairment detection system in EON. This helps to state the supremacy of the developed
system.

(a)
(b)
(c)
(d)
Figure 6. The investigation of the confusion matrix for the improved link failure detection
model in EON concerning “(a) ARPANET, (b) COST239, (c) NSFNET and (d) USNET”
6.4 Investigation of ROC for the recommended link impairment identification system in
EON over diverse classifiers
The recommended detection of the link failure system in EON is involved with the ROC
estimation and presented in Fig.7. The true and false positive rates are supported to estimate
the ROC measure of entire networks over diverse classifiers. The implemented link failure
detection system is improved by 12.5 % of LSTM, 13.2 % of RNN, 15 % of Resnet, and 14.2
% of ASPP-1DCNN appropriately from Fig.7 (a) when the false positive rate is 0.4. This
reveals that the implemented link failure detection system has better performance than the
existing link failure detection models.

(a)
(b)
(c)
(d)
Figure 7. The experiment of ROC for the improved link failure detection model in EON over
multiple existing classifiers concerning “(a) ARPANET, (b) COST239, (c) NSFNET and (d)
USNET”
6.5 The convergence validation of the developed IPBBM in the link failure detection
model in EON over various existing algorithms
To validate the improved IPBBM model the convergence analysis is conducted over multiple
existing algorithms and depicted in Fig.8. In this estimation, the IPBBM’s convergence is
estimated with the number of iterations. The convergence of the implemented IPBBM is
enhanced by 17.3 % of MAO-ASPP-1DCNN-AM, 17.4 % of TSO-ASPP-1DCNN-AM, 17.3
% of BSO-ASPP-1DCNN-AM, and 17.1 % of BMO-ASPP-1DCNN-AM appropriately in

Fig.8 (b) with the iteration count is 20. These validation solutions assured that the developed
IPBBM has a better effect than the other models.
(a)
(b)
(c)
(d)
Figure 8. The convergence estimation of the improved link failure detection model in EON
over distinct optimization algorithms regarding “(a) ARPANET, (b) COST239, (c) NSFNET
and (d) USNET”
6.6 The statistical investigation of the implemented IPBBM in the link failure detection
model in EON over various existing algorithms
The implemented IPBBM is processed with the statistical metrics to measure the
functionality over multiple existing systems for four networks and it is represented in Table
2. The metrics such as “standard deviation, median, mean, worst, and best are considered for
this process. From Table 2, the best measure of the suggested IPBBM is developed by 79.6 %

of MAO-ASPP-1DCNN-AM, 24 % of TSO-ASPP-1DCNN-AM, 79.9 % of BSO-ASPP-
1DCNN-AM, and 90 % of BMO-ASPP-1DCNN-AM accordingly when taking the NSFNET.
These final reports elucidate the improved IPBBM has better efficacy over other traditional
optimization systems.
Table 2. The statistical estimation of the improved IPBBM algorithm in the developed link
failure detection in EON over diverse existing optimization algorithms
TERMS
MAO-
ASPP-
1DCNN-AM
[29]
TSO-ASPP-
1DCNN-AM
[30]
BSO-ASPP-
1DCNN-AM
[27]
BMO-ASPP-
1DCNN-AM
[28]
IPBBM-
ASPP-
1DCNN-AM
“APRANET”
“BEST”
54.31200875
54.39848654
53.84836929
54.53065308
53.76059941
“WORST”
54.7116226
55.384732
54.5103429
55.42534482
55.45743999
“MEAN”
54.39382711
54.81973343
54.04201487
54.6035907
54.13058487
“MEDIAN”
54.41175694
54.83697441
54.04775521
54.53065308
53.76059941
“STD”
0.079121479
0.138404254
0.223845766
0.209674412
0.448686298
“COST239”
“BEST”
54.65295608
54.74085844
54.41101896
54.18359031
53.91982485
“WORST”
55.2461596
55.26433971
55.69554535
55.56955082
55.18525736
“MEAN”
54.7309626
54.83091139
54.70529062
54.29756451
53.96477932
“MEDIAN”
54.65295608
54.77904733
54.55865397
54.18359031
53.91982485
“STD”
0.133338053
0.119393072
0.320881082
0.247768514
0.178704434
“NSFNET”
“BEST”
54.40635901
54.18183479
54.40062239
54.4612575
53.97674305
“WORST”
54.81635097
54.67453171
55.54964897
55.14687359
54.81037639
“MEAN”
54.59495531
54.35508171
54.48689066
54.60010036
54.26583732
“MEDIAN”
54.40635901
54.18183479
54.40062239
54.55553905
53.97674305
“STD”
0.204338941
0.213977337
0.237321146
0.206068914
0.34278352
“USNET”
“BEST”
54.53432724
54.26771897
54.29159889
54.44174569
54.13948178
“WORST”
54.98872086
54.87472605
55.08422691
55.23854496
55.81054617
“MEAN”
54.68474263
54.315623
54.40256681
54.56469469
54.7704954
“MEDIAN”
54.61727526
54.26771897
54.29159889
54.44174569
54.93609777
“STD”
0.161825595
0.124675422
0.275031642
0.228256564
0.354327241

6.7 The K-fold validation of designed link impairment recognition in EON over diverse
algorithms and classifiers
The K-fold estimation for the implemented link failure detection model in EON over various
optimization models and classifiers and then illustrated in Fig. 9 and Fig.10 accordingly.
When considering Fig.9 (a), with a K-fold value is 2, the accuracy of the ARPANET in the
developed link failure detection model is developed by 45.8 % of MAO-ASPP-1DCNN-AM,
46.2 % of TSO-ASPP-1DCNN-AM, 46.9 % of BSO-ASPP-1DCNN-AM, and 47.3 % of
BMO-ASPP-1DCNN-AM appropriately. Thus, the improved link failure detection model
proved its efficacy.
(a)
(b)
(c)
(d)

Figure 9. The K-fold evaluation of the recommended link failure detection model in EON
over various optimization algorithms concerning “(a) ARPANET, (b) COST239, (c)
NSFNET and (d) USNET”
(a)
(b)
(c)
(d)
Figure 10. The K-fold evaluation of the recommended link failure detection model in EON
over various classifiers concerning “(a) ARPANET, (b) COST239, (c) NSFNET and (d)
USNET”
6.8 The performance estimation of the optimal routing approach in EON over various
existing algorithms
Fig.11 shows the overall functionality of the various networks in the optimal routing
approach in EON over diverse optimization algorithms. From Fig.11 (b), the distance of the

COS239 network in the optimal routing approach in EON is reduced by 31.8 % of MAO-
ASPP-1DCNN-AM, 28.2 % of TSO-ASPP-1DCNN-AM, 24.2 % of BSO-ASPP-1DCNN-
AM, and 25 % of BMO-ASPP-1DCNN-AM correspondingly. This outcome confirmed that
the developed IPBBM-ASPP-1DCNN-AM system accomplished greater efficacy.
(a)
(b)
(c)
(d)
(e)
(f)

(g)
Figure 11. The performance estimation of the recommended optimal routing approach with
several networks in EON over various optimization algorithms concerning “(a) Blocking
probability, (b) Distance, (c) Latency, (d) Link utilization (e) Load, (f) throughput and (g)
PDR”
6.9 The final estimation of the designed link failure detection model in EON over
various existing algorithms and classifiers
Table 3 and Table 4 elucidate the functionality of the improved IPBBM-ASPP-1DCNN-AM
system over multiple optimization models and classifiers accordingly for the certain
networks. The functionality factors are utilized to validate the performance of the improved
link failure detection system in EON. From Table 4, the accuracy of the USNET in the
IPBBM-ASPP-1DCNN-AM model is improved by 9.7 % of LSTM, 10.2 % of RNN, 4.7 %
of Resnet, and 8.6 % of ASPP-1DCNN respectively. This numerical experiment revealed that
the improved link failure detection in EON attained better execution rates over other existing
systems.
Table 3. The comparative estimation of the developed link failure detection system in EON
over traditional optimization systems for the several networks
TERMS
MAO-
TSO-ASPP-
BSO-ASPP-
BMO-
IPBBM-

ASPP-
1DCNN-AM
[29]
1DCNN-AM
[30]
1DCNN-AM
[27]
ASPP-
1DCNN-AM
[28]
ASPP-
1DCNN-AM
“ARPANET”
“Accuracy”
93.58228561
94.44548696
95.85288047
96.60349034
97.22274348
“Sensitivity”
93.56659142
94.46952596
95.86155004
96.61399549
97.25357412
“Specificity”
93.59790341
94.42156496
95.84425309
96.59303632
97.1920629
“Precision”
93.56659142
94.39849624
95.82549831
96.57766077
97.18045113
“FPR”
6.402096593
5.578435043
4.155746911
3.406963684
2.807937102
“FNR”
6.433408578
5.530474041
4.138449962
3.386004515
2.746425884
“NPV”
93.59790341
94.42156496
95.84425309
96.59303632
97.1920629
“FDR”
6.433408578
5.601503759
4.174501692
3.422339225
2.819548872
“F1-Score”
93.56659142
94.43399774
95.84352078
96.59582471
97.21699887
“MCC”
0.871644948
0.888909532
0.917057256
0.93206953
0.944454907
“COST239”
“Accuracy”
93.62229517
94.35698176
95.87994909
96.65929747
97.23990085
“Sensitivity”
93.6285867
94.33256104
95.8964534
96.67171627
97.23756906
“Specificity”
93.6159735
94.38151945
95.86336572
96.64681918
97.24224381
“Precision”
93.64527629
94.40406653
95.88363701
96.66310256
97.25490196
“FPR”
6.384026503
5.618480548
4.136634284
3.353180821
2.757756189
“FNR”
6.371413295
5.667438959
4.103546605
3.328283728
2.762430939
“NPV”
93.6159735
94.38151945
95.86336572
96.64681918
97.24224381
“FDR”
6.354723708
5.595933473
4.116362988
3.336897443
2.745098039
“F1-Score”
93.63693075
94.36830024
95.89004478
96.66740922
97.24623474
“MCC”
0.872445244
0.887139451
0.917598493
0.933185557
0.94479774
“NSFNET”
“Accuracy”
93.65692743
94.35438266
95.86239397
96.65409991
97.24316682
“Sensitivity”
93.68224299
94.3364486
95.87850467
96.64485981
97.25233645
“Specificity”
93.63117871
94.37262357
95.8460076
96.6634981
97.2338403
“Precision”
93.73480456
94.46004118
95.91436051
96.71717172
97.27961111
“FPR”
6.368821293
5.627376426
4.153992395
3.336501901
2.766159696
“FNR”
6.317757009
5.663551402
4.121495327
3.355140187
2.747663551
“NPV”
93.63117871
94.37262357
95.8460076
96.6634981
97.2338403
“FDR”
6.265195437
5.539958825
4.085639491
3.282828283
2.720388894
“F1-Score”
93.70851641
94.39820443
95.89642924
96.68100224
97.26597187
“MCC”
0.873130168
0.887081564
0.917242254
0.933077876
0.944859539
“USNET”
“Accuracy”
93.63470266
94.38321448
95.87434432
96.64053751
97.24759828
“Sensitivity”
93.59867722
94.41360576
95.85449392
96.63399079
97.24813984
“Specificity”
93.67058824
94.35294118
95.89411765
96.64705882
97.24705882
“Precision”
93.64291622
94.33561482
95.87714117
96.63399079
97.23665564
“FPR”
6.329411765
5.647058824
4.105882353
3.352941176
2.752941176
“FNR”
6.401322783
5.586394236
4.145506082
3.366009212
2.751860163
“NPV”
93.67058824
94.35294118
95.89411765
96.64705882
97.24705882
“FDR”
6.357083776
5.664385178
4.12285883
3.366009212
2.763344355
“F1-Score”
93.62079149
94.37459418
95.86581621
96.63399079
97.2423974
“MCC”
0.872693552
0.887664347
0.917486562
0.932810496
0.944951777

Table 4. The comparative estimation of the developed link failure detection system in EON
over traditional classifiers for the several networks
TERMS
LSTM [31]
RNN [32]
Resnet [33]
ASPP-
1DCNN[26][25]
IPBBM-
ASPP-
1DCNN-
AM
“ARPANET”
“Accuracy”
87.85888534
87.12704072
92.64402327
88.6845562
97.22274348
“Sensitivity”
87.7351392
87.20842739
92.58841234
88.71331828
97.25357412
“Specificity”
87.9820292
87.04605017
92.69936353
88.65593411
97.1920629
“Precision”
87.90049001
87.01201201
92.65813253
88.61330327
97.18045113
“FPR”
12.0179708
12.95394983
7.300636466
11.34406589
2.807937102
“FNR”
12.2648608
12.79157261
7.41158766
11.28668172
2.746425884
“NPV”
87.9820292
87.04605017
92.69936353
88.65593411
97.1920629
“FDR”
12.09950999
12.98798799
7.34186747
11.38669673
2.819548872
“F1-Score”
87.81773677
87.11010898
92.62325932
88.66328257
97.21699887
“MCC”
0.757176483
0.742542579
0.852879561
0.773690889
0.944454907
“COST239”
“Accuracy”
87.84305844
87.29595141
92.6620665
88.85911436
97.23990085
“Sensitivity”
87.88094814
87.27499554
92.60381394
88.82997683
97.23756906
“Specificity”
87.80498724
87.31700766
92.72059811
88.88839146
97.24224381
“Precision”
87.86528867
87.36452433
92.74431058
88.92903341
97.25490196
“FPR”
12.19501276
12.68299234
7.279401889
11.11160854
2.757756189
“FNR”
12.11905186
12.72500446
7.396186063
11.17002317
2.762430939
“NPV”
87.80498724
87.31700766
92.72059811
88.88839146
97.24224381
“FDR”
12.13471133
12.63547567
7.255689424
11.07096659
2.745098039
“F1-Score”
87.8731177
87.31973699
92.67400901
88.87947752
97.24623474
“MCC”
0.756859689
0.745918595
0.853242008
0.777182094
0.94479774
“NSFNET”
“Accuracy”
87.80867107
87.36098021
92.667295
88.92082941
97.24316682
“Sensitivity”
87.73831776
87.35514019
92.71028037
88.92523364
97.25233645
“Specificity”
87.88022814
87.36692015
92.62357414
88.91634981
97.2338403
“Precision”
88.0427647
87.55151742
92.74495138
89.08341916
97.27961111
“FPR”
12.11977186
12.63307985
7.376425856
11.08365019
2.766159696
“FNR”
12.26168224
12.64485981
7.289719626
11.07476636
2.747663551
“NPV”
87.88022814
87.36692015
92.62357414
88.91634981
97.2338403
“FDR”
11.9572353
12.44848258
7.255048616
10.91658084
2.720388894
“F1-Score”
87.89027758
87.45321856
92.72761264
89.00425612
97.26597187
“MCC”
0.756167688
0.747208177
0.853335877
0.778405258
0.944859539
“USNET”
“Accuracy”
87.77627159
87.25761773
92.60918253
88.81947309
97.24759828
“Sensitivity”
87.70520846
87.22097555
92.6892642
88.75634818
97.24813984
“Specificity”
87.84705882
87.29411765
92.52941176
88.88235294
97.24705882
“Precision”
87.78815463
87.24158299
92.51444065
88.82978723
97.23665564
“FPR”
12.15294118
12.70588235
7.470588235
11.11764706
2.752941176
“FNR”
12.29479154
12.77902445
7.310735798
11.24365182
2.751860163
“NPV”
87.84705882
87.29411765
92.52941176
88.88235294
97.24705882

“FDR”
12.21184537
12.75841701
7.485559354
11.17021277
2.763344355
“F1-Score”
87.74666194
87.23127805
92.60176991
88.79305252
97.2423974
“MCC”
0.755524394
0.745151294
0.852185149
0.776388521
0.944951777
7. Conclusion
The identification of the link failure system in EON was developed to forecast the link
impairment. To achieve this goal, a technique was suggested employing the hybrid heuristic
model. In the starting stage, the important data was collected and given to the link
impairment detection approach. This technique was called ASPP-1DCNN-AM in that
specific hyper-parameters were tuned by the approach called IPBBM. After detecting the link
impairment, the system demanded to locate the optimal routing for better transmission. In
this, the optimal path was recognized by employing the improved IPBBM. At last, the
formulation was conducted with the diverse metrics and contrasted with the conventional
systems. The improved link impairment detection system in COS239 attained higher
accuracy by 17.7 % of MAO-ASPP-1DCNN-AM, 17.8 % of TSO-ASPP-1DCNN-AM, 18.1
% of BSO-ASPP-1DCNN-AM, and 18.3 % of BMO-ASPP-1DCNN-AM appropriately when
taking the K-fold value as 5. Therefore, the implemented system illustrated that was attained
the good recognition solutions to make the data transmission effectively.
Conflict of Interest: The authors declare no conflict of interest
Data Availability Statement:
No new data were generated or analysed in support of this research.
Funding: This research did not receive any specific funding
Author Contribution: All authors have made substantial contributions to conception and
design, revising the manuscript, and the final approval of the version to be published. Also,
all authors agreed to be accountable for all aspects of the work in ensuring that questions

related to the accuracy or integrity of any part of the work are appropriately investigated and
resolved.
References
[1] A. Mitra, D. Ives, A. Lord, S. Savory, S. Kar, and P. Wright, "Network equipment and
their procurement strategy for high capacity elastic optical networks," Journal of Optical
Communications and Networking, vol. 8, no. 7, pp. A201-A211, July 2016.
[2] P. Papanikolaou, K. Christodoulopoulos, and E. Varvarigos, "Optimization techniques for
incremental planning of multilayer elastic optical networks, "IEEE Journal of Optical
Communications and Networking, vol. 10, no. 3, pp. 183-194, March 2018.
[3] M. Hadi and M. R. Pakravan, "Resource allocation for elastic optical networks using
geometric optimization," IEEE Journal of Optical Communications and Networking, vol.
9, no. 10, pp. 889-899, Oct. 2017.
[4] T. Ohba, S. Arakawa and M. Murata, "Virtual network reconfiguration in elastic optical
path networks for future bandwidth allocation," IEEE Journal of Optical Communications
and Networking, vol. 8, no. 9, pp. 633-644, 1 September 2016.
[5] Y. Xiong, Y. Li, B. Zhou, R. Wang, and G. N. Rouskas, "SDN enabled restoration with
triggered precomputation in elastic optical inter-datacenter networks," IEEE Journal of
Optical Communications and Networking, vol. 10, no. 1, pp. 24-34, Jan. 2018.
[6] P. Papanikolaou, K. Christodoulopoulos, and E. Varvarigos, "Joint multi-layer
survivability techniques for IP-over-elastic-optical- networks, "IEEE Journal of Optical
Communications and Networking, vol. 9, no. 1, pp. A85-A98, Jan. 2017.
[7] S. Talebi and G. N. Rouskas, "On distance-adaptive routing and spectrum assignment in
mesh elastic optical networks," IEEE Journal of Optical Communications and
Networking, vol. 9, no. 5, pp. 456-465, May 2017.

[8] M. Yaghubi-Namaad, A. G. Rahbar, and B. Alizadeh, "Adaptive modulation and flexible
resource allocation in space-division- multiplexed elastic optical networks, "IEEE Journal
of Optical Communications and Networking, vol. 10, no. 3, pp. 240-251, March 2018.
[9] X. Wang, M. Brandt-Pearce and S. Subramaniam, "Impact of wavelength and modulation
conversion on translucent elastic optical networks using MILP, "IEEE Journal of Optical
Communications and Networking, vol. 7, no. 7, pp. 644-655, July 2015.
[10] M. Zhu, S. Zhang, Q. Sun, G. Li, B. Chen and J. Gu, "Fragmentation-aware VONE in
elastic optical networks, "IEEE Journal of Optical Communications and Networking, vol.
10, no. 9, pp. 809-822, Sept. 2018.
[11] R. Wang, S. Bidkar, F. Meng, R. Nejabati and D. Simeonidou, "Load-aware
nonlinearity estimation for elastic optical network resource optimization and
management," IEEE Journal of Optical Communications and Networking, vol. 11, no. 5,
pp. 164-178, May 2019.
[12] K. Huang, W. Ji, X. Xue, and X. Li, "Design and evaluation of elastic optical access
network based on WDM-PON and OFDM technology," IEEE Journal of Optical
Communications and Networking, vol. 7, no. 10, pp. 987-994, October 2015.
[13] P. Afsharlar, A. Deylamsalehi, J. M. Plante, J. Zhao and V. M. Vokkarane, "Routing
and spectrum assignment with delayed allocation in elastic optical networks, "IEEE
Journal of Optical Communications and Networking, vol. 9, no. 3, pp. B101-B111, March
2017.
[14] M. Hadi, M. R. Pakravan and E. Agrell, "Dynamic resource allocation in metro elastic
optical networks using Lyapunov drift optimization," IEEE Journal of Optical
Communications and Networking, vol. 11, no. 6, pp. 250-259, June 2019.
[15] B. Li, W. Lu, S. Liu, and Z. Zhu, "Deep-learning-assisted network orchestration for
on-demand and cost-effective VNF service chaining in inter-DC elastic optical networks,"

IEEE Journal of Optical Communications and Networking, vol. 10, no. 10, pp. D29-D41,
Oct. 2018.
[16] B. Wang and P. -H. Ho, "Energy-efficient routing and bandwidth allocation in
OFDM-based optical networks," IEEE Journal of Optical Communications and
Networking, vol. 8, no. 2, pp. 71-84, Feb. 2016.
[17] Bin Wu, Kwan L. Yeung, Bing Hu, Pin-Han Ho, "M2-CYCLE: An optical layer
algorithm for fast link failure detection in all-optical mesh networks", Elsevier Computer
Networks, Vol. 55, Issue 3, pp. 748-758, 21 February 2011.
[18] S. S. Ahuja, S. Ramasubramanian and M. Krunz, "Single-Link Failure Detection in
All-Optical Networks Using Monitoring Cycles and Paths," IEEE/ACM Transactions on
Networking, vol. 17, no. 4, pp. 1080-1093, Aug. 2009.
[19] K.D.R. Assis, R.C. Almeida Jr, M.J. Reed, A.F. Santos, H.A. Dinarte, D.A.R. Chaves,
H. Li, S. Yan, R. Nejabati, D. Simeonidou, "Protection by diversity in elastic optical
networks subject to single link failure", Elsevier Optical Fiber Technology, Vol. 75,
no.103208, January 2023.
[20] Neeraj Mohan, Amit Wason, Parvinder S. Sandhu, "ACO based single link failure
recovery in all optical networks," Elsevier Optik, Vol. 127, Issue 20, pp. 8469-8474,
October 2016.
[21] Vela, A.P., Ruiz, M., Fresi, F., Sambo, N., Cugini, F., Meloni, G., Potì, L., Velasco,
L. and Castoldi, P., "ER degradation detection and failure identification in elastic optical
networks," IEEE Journal of lightwave technology, vol. 35, pp.4595-4604, 2017.
[22] L. Shu, Z. Yu, Z. Wan, J. Zhang, S. Hu, and K. Xu, "Dual-Stage Soft Failure
Detection and Identification for Low-Margin Elastic Optical Network by Exploiting
Digital Spectrum Information," Journal of Lightwave Technology, vol. 38, no. 9, pp.
2669-2679, 1 May 1, 2020.

[23] Sun, Kaixuan, Zhenming Yu, Liang Shu, Zhiquan Wan, Hongyu Huang, Yi Lei, and
Kun Xu, "Digital Residual Spectrum-Based Generalized Soft Failure Detection and
Identification in Optical Networks," IEEE Transactions on Communications vol.71, pp.
324-338, 2022.
[24] Bao, Bowen, Hui Yang, Qiuyan Yao, Chao Li, Zhengjie Sun, Jie Zhang, Sheng Liu,
and Yunbo Li, "LoRB: Link-oriented resource balancing scheme for hybrid C/C+ L band
elastic optical networks," Elsevier Optical Fiber Technology, vol.74, no. 103071, 2022.
[25] Wang, S., Liu, T., Luo, K. and Yang, G., "Identification of engine faults based on
acoustic emission signals using a 1DCNN-ViT ensemble model", Measurement Science
and Technology, vol.34, no. 024007, 2022.
[26] Yuhao Wang, Binxiu Liang, Meng Ding, and Jiangyun Li, "Dense Semantic Labeling
with Atrous Spatial Pyramid Pooling and Decoder for High-Resolution Remote Sensing
Imagery", remote sensing, 2018.
[27] T. Chen, Y. Zhu, and J. Teng, "Beetle swarm optimisation for solving investment
portfolio problems," in The Journal of Engineering, vol. 2018, no. 16, pp. 1600-1605, 11
2018.
[28] Mohd HerwanSulaiman, ZurianiMustaffa, Mohd MawardiSaari, and HamdanDaniyal,
"Barnacles Mating Optimizer: A new bio-inspired algorithm for solving engineering
optimization problems", Engineering Applications of Artificial Intelligence, vol.87,
January 2020.
[29] Villuendas-Rey, Y., Velázquez-Rodríguez, J.L., Alanis-Tamez, M.D., Moreno-Ibarra,
M.A. and Yáñez-Márquez, C., "Mexican axolotl optimization: A novel bioinspired
heuristic", Mathematics, vol.9, no.781, 2021.

[30] Xie, L., Han, T., Zhou, H., Zhang, Z.R., Han, B., and Tang, A," Tuna swarm
optimization: a novel swarm-based metaheuristic algorithm for global optimization.
Computational intelligence and Neuroscience, pp.1-22, 2021.
[31] H. Ming, X. Chen, X. Fang, L. Zhang, C. Li, and F. Zhang, "Ultralow Complexity
Long Short-Term Memory Network for Fiber Nonlinearity Mitigation in Coherent
Optical Communication Systems," Journal of Lightwave Technology, vol. 40, no. 8, pp.
2427-2434, April 15, 2022.
[32] M. Zhai, X. Xiang, R. Zhang, N. Lv, and A. El Saddik, "Learning Optical Flow Using
Deep Dilated Residual Networks," IEEE Access, vol. 7, pp. 22566-22578, 2019.
[33] X. Ren, H. Mosavat-Jahromi, L. Cai and D. Kidston, "Spatio-Temporal Spectrum
Load Prediction Using Convolutional Neural Network and ResNet," IEEE Transactions
on Cognitive Communications and Networking, vol. 8, no. 2, pp. 502-513, June 2022.
[34] S. K. Singh and A. Jukan, "Efficient spectrum defragmentation with holding-time
awareness in elastic optical networks, "IEEE Journal of Optical Communications and
Networking, vol. 9, no. 3, pp. B78-B89, March 2017.