Algorithm Optimization and Verification of Transmission Line

Digital Twin and Physical Model

Kun Zhang

China Electric Power Construction Group, Guizhou Electric Power Design and Research Institute Co., Ltd, Guiyang,

Guizhou Province, 550008, China

Keywords: Transmission Line, Digital Twin, Physical Model, Algorithm Optimization, Validation.

Abstract: Transmission lines are the lifeline of the power grid. Regular inspection and timely treatment of potential

security risks can help to ensure the stable operation of the power network. In order to identify the key

components and security risks in transmission lines, this study is based on the actual operation of transmission

lines, put forward a target identification algorithm of combining digital twin and physical model, and the

current common YOLOX algorithm, Faster-RCNN algorithm comparison analysis, to effectively expand the

fusion algorithm feature mapping data, realize the comprehensive integration of spatial information and

channel properties and the identification of the target accurately. In view of the research results analysis,

through the transmission line digital twin and physical model combining algorithm optimization and

validation, the validation index accuracy increased 3.64%, average validation accuracy is as high as 90.71%,

identification speed only 10.37 milliseconds, loss value of 0.70711, the transmission line digital twin

combined with the physical model algorithm optimization is more practical.

1 INTRODUCTION

Today, energy supply is crucial to the country's

economic development and public well-being. China

has a vast land area, and the transmission lines

undertaking the function of power transmission often

need to pass through the mountains and inaccessible

areas (Buljak, and Buljak, 2012). The complicated

and changeable natural conditions have a great safety

impact on the daily operation of the transmission lines

and related components and facilities. In the

construction process of construction, the power

personnel usually carry out scientific and

maintenance of the transmission lines based on

regular inspection means, and take measures to

ensure the safety risks of the transmission lines to

ensure the continuity and safety of the power grid

operation (Chen, 2015).

Traditional transmission line inspection methods

include technical staff holding special instruments

along the transmission line, channel field inspection,

etc. However, the total mileage of transmission lines

in China ranks the first in the world, and the potential

damage risk of transmission lines is difficult to

predict (Csiszár, 2007). Therefore, this kind of

manual inspection mode has problems such as low

efficiency, time-consuming and laborious, rising cost

and security risks (Ghaedi, and Bardsiri, et al. 2023).

With the progress of technology in the field of

hardware, unmanned aerial vehicles are gradually

used for the regular monitoring of power transmission

lines (Li, and Li, 2014). The operator uses advanced

equipment such as UAV to visually monitor the key

parts of the transmission line and send the obtained

model data to the central server (Lin, and Zhang, et

al. 2017). Although this method reduces the

transmission consumption and potential risks of

power resources to a certain extent, it still faces a

problem (Trojovska, and Dehghani, et al. 2022).:

directly examining a large number of model data will

inevitably lead to omissions or errors, and the

inspection results will be affected by the personal

judgment of the reviewer, so the inspection data will

be changed (Vidyasagar, and Vidyasagar, 2020). In

view of the problem of low efficiency of transmission

line inspection in the power system, many power

workers consider the application of digital technology

in this field, that is, to use digital equipment or

technology to obtain clear image data, and to use

specific model processing technology for in-depth

analysis (Yang, 2011).

Zhang, K.

Algorithm Optimization and Veriﬁcation of Transmission Line Digital Twin and Physical Model.

DOI: 10.5220/0013547700004664

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 1, pages 505-510

ISBN: 978-989-758-763-4

505

The diagnostic technology of transmission line

component defects adopts typical visual

identification technology, The process relies on

human intervention, That is, by interpreting the

content of the model, People to manually formulate

and extract the features (Younes, 2010)., The process

is relatively complicated and tedious, And most of the

solutions are built for special circumstances, Its lack

of elasticity and universality, The challenges mainly

include: first, the detection objects of transmission

lines have different scale characteristics, Thus a

single perspective inspection may lead to the loss of

detail features, It will also make the scale difference

between the different test objects become more

obvious; Second, the variety of detection objects, For

example, for the displacement identification of the

transmission line flat pressure ring., The difficulty in

this type of object identification is the movement of

transmission line components, Instead of the device

itself, The diversity of its inspection angles increases

the difficulty of identification; In addition, multiple

test objects may have similar characteristics; Third,

the inspection background of the detected objects is

very complex, Transmission lines are often laid in

mountains, rivers and other areas with complex

natural conditions., Related external factors put

forward high adaptability and universality

requirements for the algorithm model, Such as

sunshine, haze, rainstorm and so on. In order to

effectively solve the above problems., relevant

scholars gradually calculated and put forward the

research on the optimization of transmission line

algorithm, and adopted the digital twin combination

algorithm model as the reference model. This kind of

model not only abandons the traditional anchor frame

mechanism, but also shows certain advantages in

dealing with problems such as target differences.

Therefore, the update and iteration of the algorithm

combining the transmission line digital twin and the

physical model not only maintains the real-time

performance of the hidden danger detection, but also

surpasses the traditional two-stage detection

algorithm in terms of efficiency..

2 DESCRIPTION OF THE

RELATED PROBLEMS

2.1 Digital Algorithm of the

Transmission Line

In view of the low proportion of power lines in the

model, and the targets of different types to be

inspected are significantly different in size, so in the

algorithm structure of the digital twin combination

model, the deep feature integration is carried out by

the method of feature map accumulation. Using this

kind of hybrid strategy can amplify the data contained

in each dimension, and then enrich the details

contained in the feature mapping, effectively enhance

the description power of the digital twin combined

with the model feature mapping, and have a positive

effect on the completion of multi-scale target

problems, and will not increase the number of

features, and effectively reduce the resources required

for the model operation. This model subdivides the

channel focusing step on the feature diagram of

power lines into two single-dimensional feature

extraction processes, which helps to accurately

determine the target position of power lines in

complex scenes and improve the accuracy of target

detection. Therefore, the digital algorithm of the

transmission line is constructed, as shown in formula

(1):

)(wz

),(

)(

wjx



≤≤

(1)

),( wjx

In formula (1), H represents the height

and represents the output value of the processed c-th

channel in the width w direction. According to the

above calculation procedure, the feature mapping Z

can be obtainedw. Subsequently, the encoded feature

map is combined according to the channel dimension,

compressed by the convolution kernel of 11, and the

activation function is applied to build the feature

mapping model f of the middle layer, as shown in

formula (2):

))),(((

ZZConcatFf

(2)

In formula _ (2), it represents the nonlinear

activation function, F represents the convolution

factor, and represents the merging operation of

channel splicing.

),(

ZZConcat

2.2 Algorithm Optimization of

Transmission Line Digital Twin

and Physical Model

)(wz

Based on the above transmission line digital

algorithm, the digital twin algorithm is used to

INCOFT 2025 - International Conference on Futuristic Technology

506

reconstruct the decoupled tensor at the channel level

and adjust it to the c-dimensional region to form the

eigenvector gw. As shown in formula _ _ (3):

))((

fFg

(3)

In formula _ (3), the Sigmoid activation

function represents the convolution process.

)(

By performing the weighted position

multiplication on the input feature graph X, the

position data of the feature graph is integrated, and

thus an algorithm model combining the digital twin

and the physical model of the transmission line is

formed, as shown in formula (4):

),( jiy

)()(),(),( jgigjixjiy

ccc

××=

(4)

),( jix

)(ig

)( jg

)(

In formula (4),

the total amount of model parameters, the

conventional twin feature value and the twin feature

value, and the feature map output by the model will

be transmitted to the head layer to complete the

classification regression task.

2.3 Classification and Treatment of the

Loss Value

In the digital twin combination algorithm, in order to

distinguish the loss value calculated by the model,

when dx increases, that is, the deviation between the

prediction box and the actual box is large, so the loss

value is calculated from the damage model.

)3,2,1(1 =− kiou

In the face of the situation of

large positioning deviation, it gives higher

measurement value, which is beneficial to the model

to effectively screen the samples with poor accuracy.

In view of the above problems, this study based on

the digital twin concept, build digital twin

combination algorithm of loss function model,

designed to improve the prediction accuracy of model

prediction, and the detection efficiency of good

prediction box further optimization processing,

prompting prediction box and the actual marking box,

as shown in the formula (5):







≥−

≤≤−

adxioub

adxiou

loss

),(1

0,1

(5)

iou

In formula (5), it represents the mapping

relationship of independent variables in the digital

twin combination algorithm, and k represents the loss

dimension.

(3)

(, , )

iji

′

χωωω

εω

→

′

= 

(6)

3 EXPERIMENTAL RESULTS

AND ANALYSIS

3.1 Analysis of the Test Results

For assessing the accuracy of improved digital twin

combination algorithm, this study design group of

control experiment, compare the digital twin

combined sample differences, based on the traditional

physical model characteristic integration, realize the

transmission line digital twin and physical model

algorithm optimization and validation, and by

evaluating the optimal training weights, the specific

experimental results as shown in Table 1.

Table 1: Results of the ablation experiments

After the optimization of the digital twin

combination algorithm, its performance is effectively

improved. The improved digital twinning binding

algorithm A has increased the verification results

from 87.07% to 89% compared to the physical

algorithm model. However, the digital twin

combination algorithm B integrates the CA module

based on the characteristics of the algorithm A, which

improves the verification results to 89.99%. The

digital twin combination algorithm C even integrates

the above three technologies, and the final

verification result reached 90.71%. The iterative

upgrade of the digital twin combination algorithm A

to the algorithm C gradually realizes the increasing

function of the algorithm model, which proves the

Algorithm Test And Verify

Time Consuming

/Ms

Digital Twinning

Combination

Algorith

90.71 10.37

Faster-Rcnn 87.07 12.23

Yolov5 76.27 13.42

Yolov4 70.60 16.06

Yolov3 61.68 25.15

Physics Model

Algorith

66.34 66.81

Algorithm Optimization and Veriﬁcation of Transmission Line Digital Twin and Physical Model

507

effectiveness of the transmission line digital twin

combination algorithm proposed in this paper.

In this paper, the current target detection

technology in the field of digital twin combination

evaluates the safety of transmission lines in the form

of verification score. In view of the detection of

transmission line components and their defects, the

accuracy and timeliness are clearly required, this

research compares different algorithm models in

terms of processing time, and the specific test results

are shown in Table 2.

Table 2: Detection results of the different algorithms

Research

ic:

Research

Methods:

Progress

(

)

Digital

construction

Mathematical

simulations,

electrical

eriments

Operational

prediction

Power deep

learning, data

anal

sis

Transmission

line twin

digital loss

calculations

Power modeling,

numerical

analysis

Twin

maintenance

optimization

Operations

Research and

Decision Tree

Analysis for

Trans

ortation

Among the identification targets of the above

algorithm, different algorithm models show different

levels in terms of accuracy, among which the average

verification accuracy of the digital twin combination

algorithm is as high as 90.71%; in the calculation of

the model recognition speed, the most prominent

performance is the digital twin combination

algorithm, and the recognition speed is only 10.37 ms.

In this paper, the digital twin combination algorithm

model has made special adjustments to the

transmission line architecture to make the algorithm

model more streamlined and accurate.

In order to comprehensively check the

identification efficiency of various algorithm models,

the detection results are displayed graphically, as

shown in Figure 1. After observing the graphical data,

it can be seen that the digital twin combination

algorithm developed in this study has better

identification ability for closely distributed

transmission line targets compared with the

traditional physical algorithm model. Although

Faster-RCNN, as a two-step recognition algorithm,

performs well in missing targets, it brings the problem

of false identification, that is, the same target is

repeatedly identified many times under the same

environment. According to comprehensive analysis,

the digital twin combination algorithm of

transmission line has higher verification accuracy,

faster recognition speed and lower error value.

1Figure 1: Visual results of the different algorithms

3.2 Verification of Generalization

Capability

In the detection of transmission lines, it may

encounter interference from various external

conditions, such as unclear images, exposure

problems caused by overbright light, etc. In view of

this, this study takes the above external factors into

consideration, and the disturbed algorithm model is

analyzed and realizes visual processing, so as to test

the applicability and generalization ability of the

proposed algorithm. As shown in Figure 2.

2 Figure 2: Detection results in the model-based fuzzy state

As shown in Figure 2, it is clearly observed that

the digital twin combination algorithm proposed in

this study presents robust recognition performance

despite objective disturbances such as overexposure

and picture blur. In addition, other algorithms show a

INCOFT 2025 - International Conference on Futuristic Technology

508

certain degree of detection bias, such as the Faster-

RCNN algorithm misidentifies the balance ring as the

balance ring itself; in the fuzzy model scenario, the

insulator damage is not effectively detected, but the

digital twin combination algorithm accurately

identifies, so it has high identification accuracy.

Figure 3: Detection results under haze interference state

Under the same conditions, for the identification

targets of different algorithms, the excellent detection

results shown by the digital twin combination

algorithm also reflect the powerful performance of

the model. As shown in Figure 3, the identification

ability of different algorithms has obvious differences

in the environment of poor light or haze interference

environment. In contrast, YOLOv5 and Faster-

RCNN algorithms have a certain degree of

misdetection, while the digital twin combination

algorithm shows a relatively excellent detection

efficiency in the identification performance. As

shown in the Figure 4.

Figure 4: Detection results of different defects in the same

environment

In order to further verify the applicability and

generalization ability of different algorithms in the

data set, a new region was selected for comparative

analysis, and applied to the identification target of

transmission lines collected from other regions. The

final measurement results are basically the same, as

shown in Figure 5.

Figure 5: Identification of the targets in another region

As shown in the figure, the digital twin

combination algorithm proposed in this paper is

excellent, and no omission or misdetection has been

found. Based on the above analysis, for the

comparative analysis of different algorithms, the

detection effect of YOLOv4 is relatively poor, failed

to identify any detection target, other algorithms

failed to identify the shock hammer target, only the

digital twin combination algorithm detected the shock

hammer target. In addition, Faster-RCNN algorithm

does not perform well in detecting intensive targets,

and the problem that the same target is frequently

identified multiple times is more serious, but the

digital twin combination algorithm can accurately

detect, so it is further proved that the digital twin

combination algorithm has strong applicability and

generalization ability.

3.3 Calculation and Analysis of the

Loss Value

In the above digital twin combination algorithm, the

k value is set as 2 according to experience, and the b

value is calculated through the digital twin

combination algorithm, and moderate approximate

calculation is implemented. Finally, the loss value of

the digital twin combination algorithm is 0.70711.

The specific calculation results are shown in Figure

Figure 6: Loss values of the digital twin-binding algorithm

Algorithm Optimization and Veriﬁcation of Transmission Line Digital Twin and Physical Model

509

4 CONCLUSIONS

To sum up, this study for the core components of

transmission lines and safety hidden trouble accurate

detection, based on the actual operation situation, put

forward a transmission line digital twin combined

with physical model of target identification

algorithm, and with the current commonly used

YOLOX algorithm, Faster-RCNN algorithm

comparative analysis, through the integration of local

characteristic analysis, accurate comparison of

different algorithms and generalization ability, loss

value. By classification management and index

comparison, through the transmission line digital

twin and physical model algorithm optimization and

validation, the accuracy increased by 3.64%, average

validation accuracy of 90.71%, identification speed

only 10.37 milliseconds, loss value of 0.70711, in

different algorithms of various index in the highest,

the transmission line digital twin and physical model

algorithm combined positive influence on

transmission line detection.

ACKNOWLEDGEMENTS

Research and demonstration of full lifecycle digital

application of power grid engineering based on

GIM+GIS+IoT data fusion

REFERENCES

Buljak, V., & Buljak, V. (2012). Optimization algorithms.

Chen, P. (2015). Optimization algorithms Full-3d seismic

waveform inversion: Theory, software and practice (pp.

311-343).

Csiszár, S., & Ieee. (2007, Sep 13-15). Optimization

algorithms. Paper presented at the International

Symposium on Logistics and Industrial Informatics,

Wildau, GERMANY.

Ghaedi, A., Bardsiri, A. K., & Shahbazzadeh, M. J. (2023).

Cat hunting optimization algorithm: A novel

optimization algorithm. Evolutionary Intelligence,

16(2), 417-438.

Li, W. Y., & Li, W. (2014). Optimization algorithms.

Lin, Z. C., Zhang, H. Y., Lin, Z., & Zhang, H. (2017).

Optimization algorithms.

Trojovska, E., Dehghani, M., & Trojovsky, P. (2022).

Zebra optimization algorithm: A new bio-inspired

optimization algorithm for solving optimization

algorithm. Ieee Access, 10, 49445-49473.

Vidyasagar, M., & Vidyasagar, M. (2020). Optimization

algorithms.

Yang, X. S. (2011). Optimization algorithms. In S. Koziel

& X. S. Yang (Eds.), Computational optimization,

methods and algorithms (Vol. 356, pp. 13-31).

Younes, L. (2010). Optimization algorithms Shapes and

diffeomorphisms (Vol. 171, pp. 395-403).

INCOFT 2025 - International Conference on Futuristic Technology

510