IEEE POWER & ENERGY SOCIETY SECTION

Received 1 February 2024, accepted 8 April 2024, date of publication 12 April 2024, date of current version 22 April 2024.
Digital Object Identifier 10.1109/ACCESS.2024.3387968

Considering the Methods of Lightning Protection

and Early Warning for Power Transmission
Lines Based on Lightning Data Analysis
HONG HUO , DEZHI WANG, HAO CHEN, CHUNLEI ZHAO, AND QI CHENG
State Grid Dandong Electric Power Supply Company, Dandong 118000, China
Corresponding author: Hong Huo ([email protected])
This work was supported by the Science and Technology Project funded by State Grid Liaoning Electric Power Company Ltd., through the
project titled ‘‘Research on Lightning Risk Assessment and Lightning Protection Configuration Optimization Technology for Transmission
Lines Based on Improved Trip Rate Calculation Method,’’ under Grant 2023YF-31.

ABSTRACT The lightning discharges generated by lightning strikes can reach several hundred kiloamperes,
causing significant electromagnetic, mechanical, and thermal effects. These effects induce extremely high
overvoltages in power transmission systems, leading to the tripping of transmission lines. Therefore, the
study of lightning protection and early warning methods is crucial for safeguarding transmission lines.
However, previous research in the realm of active lightning protection has been limited, focusing either solely
on assessing the risk of lightning-induced tripping on the power grid side or merely predicting the patterns
of lightning activity on the disaster side. This paper proposes a novel method for lightning protection and
early warning of power transmission lines based on the analysis of lightning data. It encompasses both the
risk of lightning-induced tripping of transmission lines on the grid side and the hazards of real-time lightning
activities on the disaster side. In the warning phase, the method extensively mines historical lightning location
data and, based on predictions of thunderstorm cloud movement trajectories, determines if thunderstorm
clouds are nearing tightly packed transmission corridors. It then assesses whether these clouds present a
high-risk storm or are approaching transmission lines with low lightning protection capabilities, subsequently
integrating these findings to issue a continuous lightning-induced trip warning. In a case study conducted
in a certain part of northern China, the proposed continuous lightning strike trip warning method achieved
an accuracy rate of 80% for transmission corridors of particular interest to power grid companies’ dispatch
departments, enhancing the efficiency of warnings and reducing false alarms and missed reports.

INDEX TERMS Data-driven, lightning movement trajectory prediction, high-risk thunderstorm assessment,
lightning-induced trip alarm warning.

I. INTRODUCTION and implementation of a robust lightning protection system,

Lightning strikes impacting transmission lines are known which includes line lightning rods, grounding systems, surge
to provoke disturbances within the power system [1].Such protectors, and overhead ground wires, have undergone
discharges can hit the equipment directly or hit transmission continuous refinement and gained widespread acceptance
lines originating overvoltage surges sufficiently high to cause in the global electric power industry [4], [5], [6], [7]. The
severe fails on equipment insulation and consequently to persistent endeavor to diminish the frequency of lightning
unexpected outages [2].In extreme scenarios, these incidents strikes on transmission lines and reduce the incidence
can inflict substantial financial losses on power utilities of lightning-induced outages is a key objective for sci-
and businesses reliant on electricity [3]. The development entists, engineers, and practitioners in the electric power
sector worldwide. This effort is crucial for ensuring the
The associate editor coordinating the review of this manuscript and safety and stability of the power system’s operation [8],
approving it for publication was Guangya Yang . [9], [10].
2024 The Authors. This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.
54168 For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/ VOLUME 12, 2024
H. Huo et al.: Considering the Methods of Lightning Protection and Early Warning

Numerous studies have embarked on risk analyses con- vein, Document [17] employed forecasted trajectories of
cerning lightning strike-induced trip-outs in transmission thunderclouds to predict future lightning strike zones.
lines. In particular, Document [11] utilized ground flash It employed a grid-based method to directly calculate the
density metrics to identify segments of transmission lines vul- imminent probability of lightning strikes on transmission
nerable to lightning. This study integrated these metrics with lines, thereby facilitating the prediction and early warning of
parameters characterizing line lightning faults to compute the lightning strike-induced trip-outs. Document [18] introduced
frequency of lightning strike-induced trip-outs. The findings an innovative approach, utilizing information gleaned from
confirmed the utility of ground flash density in facilitating Lightning Location Systems (LLS) and image recognition
a differentiated assessment of lightning protection efficacy technology. This methodology enabled dynamic prediction of
across various line segments. Meanwhile, Document [12] lightning strike trip-outs, demonstrating noteworthy efficacy.
embarked on a quest to elucidate the underlying causes Finally, Document [19] focused on computing the near-future
of lightning strike-induced trip-outs in transmission lines. probability of lightning strike trip-outs based on the proxim-
This endeavor entailed a comprehensive multi-dimensional ity of ongoing lightning activities. The resultant data served as
statistical analysis, taking into account variables such as the a foundational element in completing lightning strike warning
timing of lightning strikes, geographical regions, voltage tasks, representing another step forward in this critical area of
levels of the affected lines, and the nature of the trip-outs. electrical engineering research.
This analysis successfully pinpointed vulnerabilities in line This paper concurrently addresses the risk of lightning-
lightning protection systems. Focusing on a specific case, induced tripping in power transmission lines and the hazards
Document [13] examined a 220kV multi-circuit transmission posed by real-time lightning activities, aiming to enhance
line mounted on a single tower to analyze lightning faults. the efficiency of early warning systems while reducing the
The study revealed a significant correlation between ground occurrences of false alarms and missed reports. Initially,
flash density, lightning current intensity, and the incidence the study analyzes a vast array of lightning location data
of line lightning strike trip-outs, thereby elucidating the and employs lightning identification, tracking, and analysis
direct causes of concurrent trip-outs attributable to lightning algorithms to predict the movement trajectories of thunder-
strikes. In an effort to address the limitations inherent in storm clouds. A high-risk thunderstorm assessment model is
the lightning hazard distribution map methodology and the proposed based on hypothesis testing.
quantitative calculation of lightning strikes, Document [14] During the warning phase, the research firstly calculates
introduced a novel, differentiated, and multilevel lightning the predicted positions of the thunderstorm clouds’ move-
strike risk assessment method. This method was predicated ment trajectories, and then assesses whether these clouds
on ground flash density values, adjusted for a variety of pose a high risk or are approaching transmission lines with
influencing factors. However, it is noteworthy that these low lightning protection capabilities. The evaluation process
methodologies predominantly cater to lightning protection incorporates a comprehensive set of data, including the
assessments from the power grid perspective, omitting the current position and dynamics of the thunderstorm clouds,
influence of real-time lightning activities on the disaster the specific layout of transmission lines, the current status
side on the risk of trip-outs. Furthermore, prior research on of lightning protection facilities, and historical lightning
lightning strike trip-outs has predominantly been qualitative strike records. This facilitates the accurate identification of
in nature, broadly categorizing them as lightning-induced potential lightning strike risk areas and the timely issuance of
without a quantitative differentiation of the thunderstorm continuous lightning strike tripping alerts.
intensities concurrent with the lightning strike trip-outs. Finally, the proposed warning method is validated through
In the realm of researching lightning activities on the case studies in a certain part of northern China. For the
disaster side, Chinese power grid lightning protection profes- tightly-knit transmission corridors that are of significant
sionals have made substantial contributions. Document [15] interest to power grid companies, the accuracy rate of the
leveraged a Backpropagation (BP) neural network to forecast proposed continuous lightning strike tripping alert system
future lightning current amplitudes. This was achieved by reached 80%, demonstrating its potential in enhancing the
utilizing meteorological radar data, including echo intensity, safety and reliability of power grids.
echo top height, and vertically integrated liquid water content. The innovative aspects of this paper are primarily mani-
Significantly, the study incorporated the lateral distance fested in two key areas:
between thunderclouds and transmission lines into a trip-out 1.Dual Risk Integration: This research marks the first
probability distribution model, enabling the issuance of instance in the field of lightning protection studies where
lightning strike warnings based on predetermined probability both the risk of lightning-induced tripping in power grid
thresholds. Moreover, Document [16] integrated data from transmission lines and the hazard of real-time lightning
atmospheric electric field meters and lightning location activities are concurrently considered. This dual-perspective
systems. By using variations in electric field strength and integration offers a novel viewpoint and methodology for
the proximity of lightning strikes as criteria, the study research in electrical grid lightning protection and early
successfully implemented a tiered lightning warning system warning systems. Unlike previous studies that typically focus
specifically for the Guangzhou power grid. In a similar on a single aspect of risk, this paper simultaneously analyzes

VOLUME 12, 2024 54169

 H. Huo et al.: Considering the Methods of Lightning Protection and Early Warning

the conditions of transmission lines on the grid side and are typically concentrated within an hour. Consequently,
the patterns of lightning activities on the disaster side. This the selection of a one-hour time window for this study
approach allows for a more comprehensive assessment of is grounded in the observed characteristics of lightning
the impact of lightning on power grids, thereby significantly activities. This duration is chosen to encompass the majority
enhancing the accuracy and practicality of early warning of the time frame in which lightning strike events occur.
systems.
2.Comprehensive Data Analysis and High-Risk Thun-
derstorm Assessment Model: Innovatively utilizing big B. LIGHTNING DETECTION, TRACKING, AND ANALYSIS
data analysis techniques, this study significantly improves ALGORITHMS
the accuracy of predicting thunderstorm cloud movement In order to monitor and analyze the movement trajectories
trajectories by analyzing extensive lightning location data of thunderstorm clouds and understand the patterns of
and integrating lightning identification, tracking, and analysis thunderstorm cloud motion, it is imperative to classify
algorithms. Moreover, based on hypothesis testing methods, historical lightning location data into their respective thun-
a high-risk thunderstorm assessment model is proposed. This derstorm clouds. In this paper, the authors have proposed
model comprehensively considers the characteristics of the a lightning identification, tracking, and analysis algorithm,
thunderstorm clouds themselves, along with various factors as described in [21], which accomplishes the aforementioned
such as geography and meteorology, resulting in more precise classification task. The basic approach is outlined as follows:
prediction outcomes. 1) Identification. The algorithm employs the Density-
Based Spatial Clustering of Applications with Noise
II. THE RESEARCH FOUNDATION OF THE EARLY (DBSCAN) method [22], [23], [24] to perform density
WARNING METHOD FOR CONTINUOUS LIGHTNING clustering of lightning strike points every 5 minutes within
STRIKE TRIPPING IN TIGHTLY SPACED the study area, based on their geographical coordinates.
TRANSMISSION CORRIDORS Lightning points classified in the same cluster are generally
The focus of this paper is on the continuous tripping of geographically close and are marked as belonging to the
tightly spaced transmission corridors, which is a key concern same thundercloud. The DBSCAN clustering algorithm is
for the dispatch departments of power grid companies. widely used in analyzing lightning location data and can
Moreover, the foundation of the lightning strike tripping identify clusters of various shapes. In tracking thunderclouds,
early warning research presented in this paper is based it is crucial to consider both the horizontal scale of
on analyzing the movement patterns of thunderclouds from the thunderstorm cell and its movement distance within
extensive historical lightning location data, which is then used 5 minutes. Typically, the horizontal range of a thunderstorm
to predict the real-time movement positions of thunderclouds. cell varies from a few kilometers to 20 kilometers, with
an average movement speed of about 16.4 m/s. Hence,
A. CONTINUOUS TRIPPING IN TIGHTLY SPACED a thunderstorm can move approximately 5 km within
TRANSMISSION CORRIDORS 5 minutes. Based on the horizontal scale and movement
Tightly spaced transmission corridors refer to sections of distance of the storm, the search neighborhood for the
power transmission lines clustered in narrow areas with mul- DBSCAN algorithm is set to 25 km. Additionally, following
tiple lines arranged closely together. Following the definition the parameter settings obtained through multiple experiments
proposed in reference [20], tightly spaced transmission corri- in [25], the minimum number of lightning strikes contained
dors are defined as: 1) comprising three or more transmission within a thundercloud in 5 minutes is set to 10. Furthermore,
lines; 2) with neighboring lines having a center-to-center the mean latitude and longitude of the lightning points within
distance not exceeding 600 meters; 3) each transmission line the same thundercloud are taken as the centroid of the
having at least 15 consecutive towers within the corridor. thundercloud. The choice of 5 minutes as the basic time
Senior experts in the field of lightning protection note that the unit for analyzing thunderclouds is made considering that the
lightning protection for transmission corridors with closely lightning range of a thundercloud typically does not expand
arranged lines requires further in-depth study. Tightly spaced further after 5 minutes of continuous activity.
transmission lines often originate from the same power 2) Tracking. Based on the concept of Moving Clusters
source or supply electricity to the same load. Continuous or in Spatio-Temporal (MCST) data [26], in the temporal
simultaneous lightning-induced tripping can lead to severe dimension, thundercloud centroids that are close in time
consequences, such as disruption of power generation or and geographic location are connected to form a complete
load supply, significantly threatening the reliability of power thundercloud trajectory. The criterion for judging proximity
supply. Given these concerns, the focus of this paper is on is set such that two centroids are within 25 km of each other.
the early warning of continuous lightning strike tripping in At the same time, the merging and splitting of thunderclouds
tightly spaced transmission corridors. Specifically, this study are also considered.
investigates the likelihood of subsequent lightning-induced 3) Analysis. Dense lightning strike points are allocated to
trippings within 1 hour following an initial lightning-induced different thunderclouds. By connecting the centroids of the
trip in such corridors.Based on historical data and statistical same thundercloud, the spatio-temporal movement trajectory
analysis, it has been determined that lightning activities of the thundercloud is obtained, which can be applied

54170 VOLUME 12, 2024

H. Huo et al.: Considering the Methods of Lightning Protection and Early Warning

and fit the residuals of predictions, exhibiting a strong

capability for nonlinear approximation.
Compared to other machine learning predictive methods,
including Gradient Boosting Decision Tree (GBDT) and
Light Gradient Boosting Machine (LightGBM), XGBoost
(Extreme Gradient Boosting) has the following four advan-
tages in solving regression prediction problems [31], [32]:
1) XGBoost performs a second-order Taylor expansion of
the objective function, which aids in faster and more accurate
gradient descent, whereas GBDT only uses first-order
derivatives in optimization.
2) XGBoost employs pre-sorting before iteration and
traverses to select the optimal splitting point. In contrast,
although LightGBM uses a histogram-based algorithm,
FIGURE 1. Lightning detection, tracking, and analysis algorithm schematic
diagram.
which requires less memory and has lower complexity for
data splitting, it struggles to find the most accurate data split
points.
3) XGBoost, similar to the Random Forest algorithm,
in the thundercloud movement prediction model discussed supports column and row sampling. This approach can reduce
in the following section. Additionally, discrete, sporadic computational load and the risk of overfitting.
lightning strikes, which account for less than 30% of all 4) XGBoost introduces regularization in the objective
lightning strikes, are of less concern in this study focusing function. It regularizes the structure of trees to reduce the
on continuous lightning strike tripping. The impact of these model’s variance and introduces shrinkage in each step to
sporadic strikes will be further analyzed in the case study avoid overfitting. This reduces the influence of individual
section of this paper. Since sporadic lightning strikes exhibit trees on the outcome and enhances the model’s generalization
less regularity, the installation of atmospheric electric field ability.
instruments on towers is suggested for their prevention. The present study selects the XGBoost method as the
Taking the lightning data from 17:16 to 17:30 on a specific predictive approach for thunderstorm movement trajectories.
day in 2022 as an example, the results of the lightning Additionally, traditional neural network methods and deep
detection, tracking, and analysis are illustrated in Figure 1. learning approaches are included for comparison, further
demonstrating the advantages of XGBoost in predicting
C. THUNDERCLOUD MOVEMENT TRAJECTORY thundercloud movement trajectories.
PREDICTION MODEL The following section introduces the model and parameters
For the early warning of continuous lightning strike tripping, of XGBoost. For a dataset D with n samples and each sample
accurately predicting the next movement position of a having m features, XGBoost predicts the output ŷi using K
thundercloud can significantly enhance the efficiency of the additive functions fk (xi ).
warning system. Building upon the obtained thundercloud
D = {(xi , yi )} |D| = n, xi ∈ Rm , yi ∈ R


centroid movement trajectories, this paper employs Extreme (1)
Gradient Boosting (XGBoost) to predict the movement K
X
trajectory of thundercloud centroids [27]. XGBoost is an ŷi = fk (xi ), fk ∈ F (2)
ensemble machine learning method based on decision trees, k=1
widely used for solving predictive problems. In machine
In the formula, F represents the space of decision trees.
learning competitions, XGBoost has become the method of
To learn the set of additive functions used in the model,
choice for many winning teams.
XGBoost minimizes a regularized objective I.
Existing methods for predicting thundercloud trajectory,
including linear fitting [28], nonlinear fitting methods [29], X X
I= l(ŷi , yi ) + (fk ) (3)
and the Holt two-parameter linear exponential smoothing
i k
method [30], all utilize real-time lightning location data 1
to extrapolate the movement trajectory of thunderclouds. (fk ) = γ T + ∥ω∥ λ2 (4)
2
However, the patterns of historical thundercloud movement
trajectories have yet to be integrated into these trajectory In the formula: l is a differentiable convex loss function used
prediction models. Since thundercloud movement trajectories to measure the difference between the actual value yi and
are typically nonlinear, a method with strong nonlinear the predicted value ŷi ;  is an additional regularization term
approximation capabilities is required. Therefore, this paper used to penalize the complexity of the model, thus avoiding
introduces a machine learning approach (XGBoost) to learn overfitting; γ and λ are penalty coefficients, which can be
the patterns of historical thundercloud movement trajectories. adjusted according to requirements in practical applications;
The XGBoost method continually decomposes trees to learn ω represents the weights of the leaves of the decision tree.

VOLUME 12, 2024 54171

 H. Huo et al.: Considering the Methods of Lightning Protection and Early Warning

The core concept of XGBoost is the continuous splitting of TABLE 1. Comparison of predictive performance across various methods.
features to add trees. The addition of trees serves to introduce
new additive functions ft (xi ) to fit the residuals from the
previous prediction.
(t) (t−1)
ŷi = ŷi + ft (xi ) (5)
(t)
In the formula: ŷi is the predicted value for instance i in
(t−1)
the t round of iteration; ŷi is the predicted value from the
previous round of iteration.
After substituting into equation (5), the objective function
can be rewritten as shown in equation (6). Then, the objective
function is derived using a fast optimization second-order
approximation, as shown in equation (7).
n
(t−1)
X
I (t) = l(yi , ŷi + ft (xi )) + (ft ) (6)
contains 5 centroids corresponding to the final 22 minutes
i=1
n   before the end of the thundercloud. Furthermore, under the
X 
(t−1)
 1 condition that the input of the prediction model remains
I (t) ≃ l yi , ŷi + gi ft (xi ) + hi ft2 (xi ) + (ft )
2 unchanged, the output of the model can also be the centroid
i=1
of the thundercloud for the second upcoming time period,
(7)
i.e., the next 6 to 10 minutes, or for the third upcoming time
(t−1)
gi = ∂ŷ(t−1) l(yi , ŷi ) (8) period, i.e., the next 11 to 15 minutes. As shown in Table 1,
hi =
(t−1)
∂ŷ2(t−1) l(yi , ŷi ) (9) the locations of the centroids for the first, second, and third
upcoming time periods are predicted at the current moment.
In the formula: gi represents the first-oeder gradient of the Table 1 compares the performance of various existing
loss function; hi represents the second-order gradient of the thunderstorm cloud trajectory prediction methods, including
loss function. neural network and deep learning approaches, using the
To find the optimal objective function for solving regres- absolute error in the distance between the actual and predicted
sion prediction problems and avoiding overfitting, appro- positions of the cloud centroid as the performance evaluation
priate parameters must be set for XGBoost. In regression metric. For training, data from 87,029 sets corresponding to
prediction problems, key parameters of XGBoost are the thunderstorm clouds in a certain part of northern China, for
maximum depth, learning rate, and the estimated number of the years 2020-2021 were used. For testing, data from 43,272
trees, denoted as n More specifically, the maximum depth sets corresponding to the 2022 thunderstorm clouds were
refers to the depth of the tree; the deeper the tree, the more utilized. Table 1 shows the average values of the absolute
complex the model, and the higher the risk of overfitting. error in distance for the test group. The results demonstrate
The learning rate is the shrinkage step used in the updating that XGBoost can consistently improve the precision in
process. The estimated number of trees is the count of predicting the centroid position of thunderstorm clouds.
decision trees. The latter two parameters are interrelated: the Particularly, this paper is concerned with the prediction
higher the learning rate, the fewer decision trees are required. of thunderstorm cloud movement trajectories in relation to
This paper employs Bayesian optimization to automatically transmission lines. For elongated transmission lines, there
tune these three parameters. Bayesian optimization is a is a need to further improve the predictive performance
strategy for automatic parameter adjustment with fewer of thunderstorm cloud movement trajectories. To this end,
iterations and faster iteration speed. Ultimately, the real-time the paper introduces five meteorological factors that may
movement trajectory of thunderclouds is predicted using influence the movement of thunderstorm clouds into the
XGBoost, which has been trained on historical thundercloud prediction model: temperature, dew point, wind direction,
movement trajectories and optimized in terms of parameters. wind speed, and total cloud cover.The potential interac-
The input for the thundercloud movement trajectory tions or correlations among these meteorological factors
prediction model consists of four centroids corresponding to are crucial for accurately predicting the movement and
four continuous time periods of a particular thundercloud. behavior of thunderstorm clouds. Firstly, the correlation
The output of the model is the centroid of the thundercloud for between temperature and dew point is instrumental in
the next time period, specifically for the next 1 to 5 minutes. determining atmospheric humidity levels. A combination of
During the training phase of the algorithm, if a thundercloud high temperatures and high dew points typically signifies
exists for 97 minutes, it will have a total of 20 centroids, with a higher moisture content in the air, which is a critical
the last centroid corresponding only to the last 2 minutes of condition for the development of thunderstorm clouds.
lightning strikes. This thundercloud can generate 16 sets of Secondly, the close relationship between wind direction
training data; the first set contains 5 centroids corresponding and wind speed plays a pivotal role in the movement and
to the initial 25 minutes of the thundercloud, and the last set evolution of thunderstorm clouds. Wind direction influences

54172 VOLUME 12, 2024

H. Huo et al.: Considering the Methods of Lightning Protection and Early Warning

the path of cloud movement, while wind speed affects TABLE 2. Comparison of the performance of various methods in
predicting the centroid of thunderstorm cloud segments causing lightning
the velocity and dispersion of the cloud. Additionally, the strike trip-outs in transmission lines.
interaction between total cloud cover and temperature can
impact ground-level temperatures, which in turn influences
the thermodynamic conditions of thunderstorm clouds. The
effects of wind speed and direction on total cloud cover
are also significant. Strong winds might disperse cloud
cover, whereas certain wind directions can contribute to the
formation and accumulation of clouds, facilitating the genesis
of thunderstorm clouds. Lastly, the combination of dew point,
temperature, and cloud cover can lead to an increase in
cloud cover, creating conducive conditions for thunderstorm
activity. Considering these factors collectively allows for a
more comprehensive understanding of thunderstorm cloud
behavior, thereby enhancing the accuracy of predictions.
However, as the length of the input sequence increases, the
span of dependencies between sequences becomes longer,
making the gradient descent-based parameter optimization
calculations more complex and significantly increasing train- to target the centroids of thunderstorm cloud segments that
ing time. Therefore, it is necessary to select meteorological cause lightning strike trip-outs on transmission lines.
factors highly correlated with the movement of thunderstorm As indicated in Table 2, 10GBoost-dCor has further
clouds as input data.The distance correlation (dCor) method improved the prediction accuracy. Therefore, this study opts
is utilized to calculate the distance correlation coefficients for XGBoost-dCor to predict the movement trajectories of
between each meteorological factor and the centroid of thunderstorm clouds.
thunderstorm cloud segments [33], [34]. Considering that the In summary, the prediction of thunderstorm cloud move-
centroid’s position coordinates are two-dimensional, the dCor ment trajectories discussed in this section can be utilized to
method can analyze the correlation between meteorological predict the distance between the centroid of thunderstorm
factors and the movement centroid of thunderstorm clouds clouds and transmission lines. This will serve as the basis
without being limited by dimensions. Numerical fitting for triggering continuous lightning strike trip-out warnings
results indicate that dCor can effectively detect nonlinear in closely spaced transmission corridors.
or non-monotonic relationships between variables, especially
when the correlation between random variables is nonlinear. III. METHOD FOR EARLY WARNING OF CONTINUOUS
Based on the thunderstorm and meteorological data from a LIGHTNING STRIKE TRIP EVENTS IN COMPACT
certain part of northern China, for the years 2020–2021, the TRANSMISSION CORRIDORS
distance correlation coefficients between temperature, dew To prevent the hazards of lightning to transmission lines,
point, wind direction, wind speed, total cloud cover, and lightning protection management strategies primarily con-
the centroid of thunderstorm cloud segments were found to sider the following factors:
be 0.206, 0.591, 0.112, 0.099, and 0.296, respectively. In 1) Lightning activity, which is the direct cause of
this study, only the three meteorological factors with higher lightning-induced tripping in transmission lines. The mag-
distance correlation coefficients (temperature, dew point, and nitude of lightning over-voltage can reach several hundred
total cloud cover) were selected along with the centroid thousand volts, a level generally beyond the endurance of
position data as inputs for the XGBoost model. This predic- power equipment insulation.
tive approach is referred to as XGBoost-dCor.Furthermore, 2) The structural characteristics of transmission lines are
to verify the effectiveness of the selection of meteorological the inherent factors leading to lightning-induced tripping.
factors, a prediction using all five meteorological factors Lines with poor lightning protection performance are more
(temperature, dew point, total cloud cover, wind direction, susceptible to lightning strikes. Intrinsic factors affecting
and wind speed) combined with the centroid position data line lightning protection include the protection angle of the
was made using XGBoost and is denoted as XGBoost-All. lightning rod, grounding resistance, tower height, etc [35].
Additionally, a prediction method using only the centroid 3) Topographical and geomorphological characteristics are
position data as the input for XGBoost is referred to as environmental factors that lead to lightning-induced tripping
XGBoost. in transmission lines [36]. Side flashes are the main cause
Predicting the movement trajectories of thunderstorm of lightning-induced tripping in high-voltage lines, and are
clouds with respect to transmission lines is essential to gain significantly influenced by topography and geomorphology.
more predictive information before lightning-induced trip- Environmental factors affecting line lightning protection
outs occur. Consequently, in further testing the performance include topographical factors, climatic factors, and the
of various methods, the predictive output has been adjusted altitude of towers.

VOLUME 12, 2024 54173

 H. Huo et al.: Considering the Methods of Lightning Protection and Early Warning

In light of the aforementioned factors, this article TABLE 3. Mean values of each parameter and the p-value from
hypothesis testing.
introduces a high-risk thunderstorm assessment model to
determine the risk of lightning-induced tripping due to
real-time lightning activity. Additionally, the lightning pro-
tection assessment model for transmission lines presented
in literature [21] can be utilized to evaluate the risk of
lightning-induced tripping relevant to the structural charac-
teristics of the transmission lines and their topographical and
geomorphological features.

A. CONTINUOUS HIGH-RISK THUNDERSTORM

ASSESSMENT MODEL
The lightning positioning system records the location, time,
lightning current value, and number of return strokes for
each ground flash. This paper is concerned with the records
of lightning activities that cause tripping. According to the
case studies in the Electric Power Company’s lightning
protection manual, lightning activity records are sought
within a 5-minute window before and after the tripping record testing method is introduced to determine whether there
time and within a 5-kilometer radius around the faulted are significant differences in each parameter between all
line. Considering that the lightning strike points found in thunderstorm cloud segments and high-risk thunderstorm
this manner are generally not unique, this paper opts to use cloud segments.
segments of thunderstorm clouds to correspond to a particular Taking the thunderstorm clouds in a certain part of northern
tripping record. A thunderstorm cloud segment refers to China, from 2020 to 2021 as an example, the mean values
dividing a thunderstorm cloud into multiple segments on of the aforementioned five parameters for all thunderstorm
the time dimension, where each segment contains all the cloud segments and those causing tripping are shown in
lightning strike points within 5 minutes of that thunderstorm Table 3. In Table 3, the p-value is the probability of
cloud. If a thunderstorm cloud disappears before 5 minutes, hypothesis, an important indicator in inferential statistics, and
it is also recorded as a segment. serves as crucial evidence for judging the correctness of the
In researching lightning-induced tripping, the initial step original hypothesis.
involves identifying all the lightning strike points contained Considering that the distribution of each parameter does
within multiple thunderstorm cloud segments corresponding not conform to a normal distribution or other common
to the time of the tripping record. Subsequently, in terms distributions, this paper employs non-parametric tests to
of geographical location, and in conjunction with the results conduct statistical inference on the various parameters of
of traveling wave ranging, the nearest lightning strike point thunderstorm clouds.
around the approximate fault location is identified from The two-sample Kolmogorov-Smirnov (KS) test is often
the aforementioned strike points. The thunderstorm cloud used to assess the difference between the cumulative
segment to which this lightning strike point belongs is then distributions of two sample datasets [37]. Under a certain
designated as the ‘‘thunderstorm cloud segment responsible significance level α, the null hypothesis posits that there is no
for the tripping.’’ significant difference in the distributions of the two samples.
The goal of the High-Risk Thunderstorm Assessment The decision of whether there is a significant difference
Model is to identify high-risk thunderstorm cloud segments between the two samples is made based on the p-value
from all thunderstorm cloud segments, that is, the segments obtained from the KS test. Specifically, for the parameters
most likely to cause tripping. The basic data that the lightning of thunderstorm clouds in a certain part of northern China,
positioning system can provide includes the location, time, from 2020 to 2021, at a significance level of α = 0.001, the
lightning current value, and number of return strokes for each p-value for the mean number of lightning strikes of the two
lightning strike point. Accordingly, there are five parameters samples is much less than 0.001, leading to the rejection of
that can be identified for a thunderstorm cloud: 1) the number the null hypothesis. This implies, at a 0.001 error probability,
of lightning strikes contained within the thunderstorm cloud; that: first, there is a significant difference in the number of
2) the average (absolute) value of the lightning current of lightning strikes between all thunderstorm cloud segments
the strike points contained within the thunderstorm cloud; and those causing tripping; second, there is a significant
3) the average number of return strokes of the strike points difference in the maximum lightning current between all
contained within the thunderstorm cloud; 4) the maximum thunderstorm cloud segments and those causing tripping;
(absolute) value of the lightning current of the strike points third, there is a significant difference in the maximum
contained within the thunderstorm cloud; 5) the maximum number of return strokes between all thunderstorm cloud
number of return strokes of the strike points contained segments and those causing tripping. Similarly, at an error
within the thunderstorm cloud. For this purpose, a hypothesis probability of 0.001, there is no significant difference in the

54174 VOLUME 12, 2024

H. Huo et al.: Considering the Methods of Lightning Protection and Early Warning

TABLE 4. Distribution test of the minimum distance from thunderstorm

cloud centroids to compact transmission corridors.

rectangle, potentially leading to a certain degree of false

alarms (i.e., even though the compact transmission corridor
is inside the minimum bounding rectangle, there may be
no nearby lightning strike points), this paper also takes into
account the minimum distance from the centroid of the
lightning strike point set to the compact transmission corridor.
This metric reflects the average distance from the lightning
FIGURE 2. Schematic diagram of the minimum bounding rectangle for a strike points to the compact transmission corridor. If this
thunderstorm cloud segment.
metric is too large, it is not considered that the thunderstorm
cloud segment impacts the compact transmission corridor.
The paper uses a one-sided confidence interval to calculate
average lightning current between all thunderstorm cloud
the upper limit of this minimum distance under a certain
segments and those causing tripping, nor is there a significant
confidence level.
difference in the average number of return strokes.
A statistical analysis was conducted on the minimum
In summary, this paper selects the number of lightning
distance from the centroids of thunderstorm clouds that
strikes, the maximum lightning current, and the maximum
caused tripping in compact transmission corridors in a
number of return strokes as the candidates for the basis
certain part of northern China, from 2020 to 2021. Various
of high-risk thunderstorm assessment. Specifically, for the
distribution tests were performed on these distances, and the
compact transmission corridors of interest in this study,
results are presented in Table 4.
a minimum bounding rectangle is used to determine whether
At a significance level of α = 0.05, the null hypothesis
a thunderstorm cloud segment is near a compact transmission
assumes that the distribution of the minimum distance follows
corridor. The minimum bounding rectangle is defined as
a certain distribution. Following the distribution tests, the
the smallest rectangle encompassing all the lightning strike
p-values for the Poisson, uniform, and Rayleigh distribution
points of a thunderstorm cloud segment, describing the
tests are all less than 0.05, meaning the null hypothesis is
potential impact range of the thunderstorm cloud segment.
rejected, and the distribution of the minimum distance does
The solution for the minimum bounding rectangle is based on
not follow the Poisson, uniform, or Rayleigh distributions.
the approach provided in literature [38], [39], which primarily
However, the p-value for the normal distribution test is
involves calculating the convex hull of the current set of
0.1011, greater than 0.05, implying acceptance of the null
lightning strike points (the convex hull being the convex
hypothesis that the distribution of the minimum distance
polygon formed by connecting the outermost points) and then
follows a normal distribution. Under the condition that
calculating the area of the rectangle corresponding to any two
the distribution of the minimum distance follows a normal
adjacent points on the hull. The specific steps are: 1) Take
distribution, the one-sided confidence upper limit at a
the line connecting two adjacent points on the convex hull
confidence level of 0.95 is
as one side of the rectangle. 2) Find the point on the convex
X −µ d −µ
 
hull farthest from the obtained edge and draw a parallel line
0.95 ≤ P {X < d} = P <
through this point to form the second side of the rectangle. σ σ
3) Project the points on the convex hull onto the obtained d −µ
 
edge and find the two projection points that are the farthest =8 ≥ 0.95 = 8 (1.6449) (10)
σ
apart; draw a line through these two points to form the other
two sides of the rectangle. 4) Iterate over all adjacent pairs After substituting in the mean and standard deviation, the
of points on the convex hull, i.e., repeat steps 1)–3), and solution yields d=39.4km. This means that at a confidence
take the rectangle with the smallest corresponding area as the level of 0.95, the one-sided confidence upper limit for the
minimum bounding rectangle. As shown in Figure 2, taking minimum distance from the centroids of thunderstorm clouds
the thunderstorm cloud segment from June 5, 2020, 06:55:06 that caused tripping in compact transmission corridors to the
to 07:00:06 as an example, the minimum bounding rectangle compact transmission corridors is 39.4km.
of this thunderstorm cloud includes a compact transmission In summary, if the minimum bounding rectangle of a
corridor (with four transmission lines). thunderstorm cloud segment includes a compact transmission
Considering that some lightning strike points far from corridor, and the predicted position of the next centroid of this
the centroid can enlarge the area of the minimum bounding thunderstorm cloud segment is less than 39.4km away from

VOLUME 12, 2024 54175

 H. Huo et al.: Considering the Methods of Lightning Protection and Early Warning

FIGURE 3. Proportion of lightning-induced tripping corresponding to

different average lightning strike count ranges. FIGURE 4. Proportion of lightning-induced tripping corresponding to
different ranges of average maximum lightning current values.

the compact transmission corridor, it is then designated as a

‘‘thunderstorm cloud segment near the compact transmission
corridor.’’
In this paper, the input for the early warning system
consists of groups of four consecutive thunderstorm cloud
segments, and the output is the warning status of that
thunderstorm cloud for the next time period (5 minutes).
Initially, all groups of thunderstorm cloud segments in a
certain part of northern China, from 2020 to 2021 were
statistically analyzed. If the fourth thunderstorm cloud
segment in a group is identified as being near the compact
transmission corridor, the average number of lightning strikes
for the four thunderstorm cloud segments in that group is FIGURE 5. Proportion of lightning-induced tripping corresponding to
different ranges of average maximum number of return strokes.
calculated, and the group is labeled as a ‘‘thunderstorm cloud
segment group near the compact transmission corridor.’’ If a
group of thunderstorm cloud segments is identified as being
near the compact transmission corridor and the subsequent segment in a group is near a compact transmission corridor
thunderstorm cloud segment causes a lightning-induced trip, and the average number of lightning strikes for the group
then that group is also marked as a ‘‘thunderstorm cloud exceeds 300, it triggers an early warning for the thunderstorm
segment group causing lightning-induced tripping.’’ The cloud in the subsequent time period.
proportion of lightning-induced tripping shown in Figure 3
refers to the percentage of groups identified as causing B. TRANSMISSION LINE LIGHTNING PROTECTION
lightning-induced tripping within the same average number ASSESSMENT MODEL
of lightning strikes range among the groups that are near the Previous assessments of lightning-induced tripping risk
compact transmission corridor. typically qualitatively categorized such events as lightning-
Similar to the average number of lightning strikes for induced tripping, without quantitatively distinguishing the
groups of thunderstorm cloud segments, the average values intensity of the thunderstorm at the time of the tripping.
of the maximum lightning current and the maximum number Literature [21] introduced the concept of real-time lightning
of return strokes for these groups can also be calculated intensity and quantitatively analyzed this intensity based
to determine the corresponding proportions of lightning- on the proximity of lightning strikes, the magnitude of the
induced tripping. The results are respectively illustrated in lightning current, and the number of times a thunderstorm
Figures 4 and 5. crosses a transmission line. It also considered the impact of
By comparing the proportions of lightning-induced trip- lightning intensity and tower characteristics on the lightning
ping in each range in Figures 3, 4, and 5 (indicated by protection performance of transmission lines during tripping
blue triangles in the graphs), it can be concluded that events. The authors of this paper refer to the transmission
groups of thunderstorm cloud segments with an average line lightning protection assessment model proposed in the
number of lightning strikes exceeding 300 have the highest referenced literature [21].
proportion of causing tripping in compact transmission For towers with records of lightning-induced tripping, the
corridors. Therefore, this paper selects this criterion for model defines their lightning protection performance based
early warnings. Specifically, if the fourth thunderstorm cloud on the real-time lightning intensity at the time of tripping.

54176 VOLUME 12, 2024

H. Huo et al.: Considering the Methods of Lightning Protection and Early Warning

TABLE 5. Structural characteristics and their value ranges, as well as

topographical and geomorphological characteristics and their value
ranges.

For towers without tripping records, the model searches

for the most similar tower with a tripping record based on
structural and topographical features (as shown in Table 5).
The risk level of the most similar tower with a tripping
record is then assigned to the tower without a tripping record.
In summary, this approach allows for the determination of
the risk level for all towers, and consequently, the assessment
of the lightning protection risk level for the transmission
line.
The early warning system in this paper selects transmission
lines ranked in the lowest 10% in terms of lightning protection
performance as low lightning protection performance trans-
mission lines. If the fourth thunderstorm cloud segment in a
group is identified as a segment near a compact transmission
corridor, and this compact transmission corridor contains low
lightning protection performance transmission lines, then an
early warning is triggered for the thunderstorm cloud in the FIGURE 6. Continuous lightning-induced tripping warning process for
subsequent time period. high-voltage transmission corridor.

C. CONTINUOUS LIGHTNING STRIKE TRIP WARNING

PROCESS FOR COMPACT TRANSMISSION CORRIDORS
Integrating the models from the two previous sections, the cloud belongs to a high-risk category. These two conditions
process involves firstly determining if a thunderstorm cloud are independent of each other; satisfying either one can
segment is near a compact transmission corridor, and then trigger a continuous lightning strike trip warning for the com-
assessing whether the compact transmission corridor contains pact transmission corridor. The flowchart for the continuous
low lightning protection performance transmission lines. lightning strike trip warning process in compact transmission
Concurrently, it evaluates whether the current thunderstorm corridors is illustrated in Figure 6.

VOLUME 12, 2024 54177

 H. Huo et al.: Considering the Methods of Lightning Protection and Early Warning

IV. NUMERICAL CASE VALIDATION 20 minutes of the initiation of the thunderstorm cloud. Due
The model proposed in this paper has been validated through to the short duration of this event, the input data for the
a case study conducted in a specific area in the northern warning in this paper consisted of consecutive segments of
regions of China. The original lightning activity data was at least 4 thunderstorm cloud slices, meaning that predictions
sourced from lightning detection stations deployed by the could only be made at least 20 minutes after the onset
electric power company in the area, with their actual detection of the thunderstorm. This limitation is inherent in the
range covering the regions in northern China intended for warning method. Compared to meteorological radar systems
monitoring. Although there were lightning-induced power that provide large-scale warnings before the occurrence
outages in the transmission system of this area, the thunder- of thunderstorms, the use of lightning location system
storms causing these outages may not be confined to this data for thunderstorm trajectory prediction can enhance the
region alone. Consequently, we utilized lightning location accuracy of predictions. However, the prerequisite is that the
data from this region of China to track the thunderstorms, thunderstorm has been ongoing for some time, with at least
ensuring the comprehensiveness of our analysis. Based on 20 minutes of lightning location data available for prediction.
the lightning detection stations deployed in a northern region, For these two missed consecutive power outage events, the
ground lightning data were acquired, totaling over 17 million installation of atmospheric electric field instruments on the
ground lightning records from 2020 to 2022. The data transmission lines is suggested for early warning purposes.
from 2020 to 2021 were utilized for model training, while
the 2022 data were employed for case study validation. All A. EARLY WARNING PERFORMANCE VERIFICATION
structural characteristics of the transmission towers, the local METHOD
topography, climate, and line crossings around the towers, Utilizing performance evaluation criteria inspired by meteo-
as well as the outage records of the towers, were provided by rology and previous assessments of meteorological disaster
the local power grid company. Elevation data for the towers warnings for transmission lines, the performance of the
were obtained from the Shuttle Radar Topography Mission consecutive lightning-induced power outage warning method
(SRTM), and soil type data for the areas around the towers is assessed, as depicted in Table 7. In this context, NTP
were sourced from the GlobeLand land cover product, both signifies the instances in which a warning was issued and
with a spatial accuracy of 30 meters. Other topographical data consecutive power outages indeed transpired. NFP denotes
for the towers were derived through interpolation based on the occurrences when a warning was issued, but consecutive
SRTM data. Python and Matlab served as the development power outages failed to manifest. NFN accounts for scenarios
environments for this study. in which a warning was not issued, yet consecutive power
In 2022, there were a total of 145 instances of lightning- outages did occur. Finally, NTN corresponds to situations
induced power outages within the tightly integrated trans- where no warning was issued, and consecutive power outages
mission corridors in a certain part of northern China. did not take place.
Among these, 8 cases involved consecutive power outages Based on the statistical values presented in Table 7,
occurring within 1 hour of each other. For consecutive this paper employs four evaluation metrics to assess the
lightning-induced power outages occurring within the same performance of the consecutive lightning-induced power
transmission corridor, if a warning is triggered within 1 hour outage warning method. These specific evaluation metrics
after the first outage and before the second outage, it is include classification accuracy, warning accuracy, missed
considered a successful warning. Typically, the conditions detection rate, and false alarm rate. The details of each
for triggering consecutive lightning-induced power outage evaluation metric are as follows.
warnings are satisfied immediately after the first outage.
Furthermore, the warning method presented in this paper is 1) CLASSIFICATION ACCURACY, DENOTED AS ACR
based on real-time lightning location data. The time required In this context, both scenarios of ‘‘issuing a warning, and
for the lightning detection station to process and detect consecutive power outages actually occurring’’ and ‘‘not
lightning strikes at ground points is less than 1 millisecond, issuing a warning, and consecutive power outages not
and the lightning monitoring system can locate lightning occurring’’ are considered as ‘‘correct classifications.’’ ACR
strikes within seconds. However, considering that there is the proportion of the number of correct classifications
is some delay in transmitting real-time lightning location to all possible cases and is expressed by the following
data and power outage information to the data center, the formula:
warning time indicated in this paper is uniformly delayed by NTP + NTN
5 seconds. ACR = (11)
Using the warning method proposed in this paper, NTP + NFP + NFN + NTN
the subsequent power outage warning situations for these
8 events are summarized in Table 6. Among them, 2 events 2) WARNING ACCURACY, DENOTED AS AWR
did not trigger warnings. One of these events was caused Consider the scenario where a warning was issued, and
by isolated and sporadic lightning strikes. The other event consecutive power outages actually occurred as ‘‘Warning
involved consecutive power outages occurring within the first Accuracy.’’ AWR is the proportion of the number of accurate

54178 VOLUME 12, 2024

H. Huo et al.: Considering the Methods of Lightning Protection and Early Warning

TABLE 6. Consecutive lightning-induced power outage event warning TABLE 7. Performance metrics for early warning.
status.

TABLE 8. Performance of the early warning method in this paper.

3) MISSED DETECTION RATE, DENOTED AS MWR

Consider the scenario where no warning was issued, but
consecutive power outages actually occurred as ‘‘Missed
Detection.’’ MWR is the proportion of the number of missed
detections to the total number of actual consecutive power
outage occurrences, expressed by the following formula:
NFN
MWR = (13)
NTP + NFN

4) FALSE ALARM RATE, DENOTED AS FWR

Consider the scenario where a warning was issued, but
consecutive power outages did not actually occur as ‘‘False
Alarm.’’ FWR is the proportion of the number of false alarms
to the total number of actual warnings issued, expressed by
the following formula:
NFP
FWR = (14)
NTP + NFP

B. COMPARATIVE ANALYSIS OF EARLY WARNING

PERFORMANCE
To validate the effectiveness of the early warning method
proposed in this paper, a comparative analysis was conducted
with scenarios that considered only the high-risk thunder-
storm assessment model and scenarios that considered only
the lightning protection assessment model for transmission
lines. The performance of these three methods is presented
in Tables 8 to 10. It’s worth noting that, as shown in
Table 6, the 8 consecutive lightning-induced power outage
events corresponded to a total of 18 lightning-induced power
outages, including 8 initial outages and 10 subsequent
outages. In cases where two subsequent outages occurred
simultaneously after triggering a warning, it is counted as
issuing 2 warnings and is recorded as 2 actual occurrences
of consecutive power outages.
warnings to the total number of actual consecutive power The continuous power outage warning in closely moni-
outage occurrences, expressed by the following formula: tored transmission channels is the primary focus of this paper.
Therefore, it is desirable to have a lower missed detection
NTP rate, and given the relatively low total number of issued
AWR = (12)
NTP + NFN warnings, a moderate false alarm rate can be acceptable.

VOLUME 12, 2024 54179

 H. Huo et al.: Considering the Methods of Lightning Protection and Early Warning

TABLE 9. Performance of the early warning method in this paper when providing assistance to power grid companies in making
considering only the high-risk thunderstorm assessment model.
relevant control decisions.
The currently proposed method is subject to constraints
arising from geographical and meteorological factors, leading
to potential significant variances in the system’s effectiveness
across different geographical areas. The performance of the
early warning system may be influenced by the distinct
characteristics of the terrain, prevalent weather patterns, and
the frequency of lightning events in various regions. Con-
TABLE 10. Performance of the early warning method in this paper when currently, the system’s reliance on predicting the movement
considering only the lightning protection assessment model for of thunderstorm clouds is contingent upon meteorological
transmission lines.
conditions, whose variability and unpredictability could
result in inaccuracies in forecasts, leading to false alarms or
missed warnings. Furthermore, the quality and completeness
of lightning location data, historical lightning strike records,
and information regarding the layout and status of power
transmission lines are crucial for the system’s accuracy.
Insufficiencies or inaccuracies in these data sets could impair
the system’s ability to accurately predict and assess risks.
Therefore, these factors must be taken into account in
All three scenarios exhibit high classification accuracy (all practical applications to ensure the system’s efficacy and
exceeding 90%). The early warning method proposed in accuracy.
this paper has the lowest false alarm rate among them. As power grid companies start installing atmospheric
Most importantly, compared to scenarios considering only electric field instruments on transmission lines, obtaining
high-risk thunderstorms or only low lightning protection pre-warning data before ground flashes occur, this can
performance transmission lines, the proposed early warning compensate for the inherent limitations of lightning location
method has the lowest missed detection rate, while achieving systems used for early warning. In the future, it has the
an 80% warning accuracy. The high-risk thunderstorm assess- potential to further reduce the missed detection rate and
ment model and the lightning protection assessment model improve the accuracy of early warnings.
for transmission lines complement each other, effectively
improving the accuracy of the warnings. After excluding REFERENCES
consecutive power outages caused by discrete occasional [1] E. A. Mahdiraji, M. S. Amiri, and S. M. Shariatmadar, ‘‘Analysis of
lightning strikes and accounting for the inherent limitations lightning strikes on the transmission line by considering the frequency-
of lightning prediction, the proposed early warning method dependent model,’’ Quantum J. Eng., Sci. Technol., vol. 2, no. 6, pp. 12–36,
2021.
issues warnings for 100% of the continuous power outages in
[2] I. Petrovic, S. Nikolovski, H. R. Baghaee, and H. Glavaš, ‘‘Determining
closely monitored transmission channels. impact of lightning strike location on failures in transmission network
elements using fuzzy decision-making,’’ IEEE Syst. J., vol. 14, no. 2,
V. CONCLUSION pp. 2665–2675, Jun. 2020.
[3] T. Debao, F. Yuhua, and Z. Xuehui, ‘‘Statistical analysis of lightning
This paper, driven by lightning location data, proposed a disaster accidents in China from 2012 to 2017,’’ Sci. Technol. Rev., vol. 36,
continuous lightning-induced power outage early warning no. 5, pp. 42–47, 2012.
method for closely monitored transmission channels, and the [4] D. Wu and J. Wang, ‘‘Lightning protection of 10-kV distribution lines
following conclusions were drawn: by multiple breakpoints arc-extinguishing lightning protection gap,’’ IEEE
Trans. Plasma Sci., vol. 48, no. 2, pp. 531–536, Feb. 2020.
1) The paper effectively harnessed the data value of [5] D. Wu, Z. Ji, and J. Wang, ‘‘Simulation and experimental analysis of multi-
the lightning location system, utilizing a vast amount of chamber arc-quenching arresters (MCAA) for 10 kV transmission lines,’’
historical data for model training, and integrating real-time Energies, vol. 14, no. 19, p. 6185, Sep. 2021.
lightning location data to issue lightning-induced power [6] G. Sheng, Y. Qian, and L. Luo, ‘‘Key technologies and application
prospects for operation and maintenance of power equipment in new power
outage warnings. systems,’’ High Voltage Eng., vol. 47, no. 9, pp. 3072–3084, 2021.
2) The introduction of a high-risk thunderstorm model [7] A. Ozdemir and S. Ilhan, ‘‘Experimental performance analysis of con-
and the simultaneous consideration of both high-risk ventional and non-conventional lightning protection systems-preliminary
thunderstorms and low lightning protection performance results,’’ Electr. Power Syst. Res., vol. 216, Mar. 2023, Art. no. 109080.
[8] J. Zhao, C. Yu, and H. Ren, ‘‘Analysis of lightning strike fault cases in
transmission lines in the early warning method enhanced special forms of 1000kV ultra-high voltage transmission lines,’’ Electr.
the efficiency of warnings, reducing the occurrence of false Porcelain Surge Arresters, vol. 18, no. 1, pp. 107–112, 2022.
alarms and missed detection events. [9] H. Zhang, ‘‘Design and analysis of lightning protection and grounding
3) For closely monitored transmission channels of interest for overhead transmission lines,’’ Electr. Technol. Econ., no. 4, pp. 69–71,
2021.
to power grid companies, the accuracy of the proposed contin-
[10] Z. Lei and H. Liu, ‘‘Electromagnetic compatibility technology of power
uous lightning-induced power outage warning reached 80%, system automation equipment,’’ Electron. World, no. 16, pp. 208–209,
demonstrating practical value in engineering applications and 2021.

54180 VOLUME 12, 2024

H. Huo et al.: Considering the Methods of Lightning Protection and Early Warning

[11] X. Tong, J. Chen, and Q. Chen, ‘‘Research on lightning parameter [35] S. Zeng, K. Liu, and Y. Han, ‘‘Analysis of lightning faults and lightning
characteristics and application in Hainan power grid,’’ Electr. Porcelain protection measures for overhead distribution lines,’’ Develop. Innov.
Surge Arresters, no. 1, pp. 134–138, 2015. Machinery Electr. Products, vol. 36, no. 6, pp. 98–100, 2023.
[12] Z. Gao, H. Pan, and F. Zhou, ‘‘Multidimensional analysis of lightning- [36] J. Zhang, Y. Zhang, and C. Liu, ‘‘Study on atmospheric electric field
induced trip-outs in Yunnan power grid transmission lines,’’ Yunnan Electr. characteristics of lightning activity in complex terrain regions of Sichuan
Power Technol., vol. 46, no. 3, pp. 20–22, 2018. province,’’ Plateau Mountain Meteorol. Res., vol. 41, no. 4, pp. 56–60,
[13] X. Peng, L. Jin, and R. Wang, ‘‘Analysis and prevention measures of 2021.
lightning-induced simultaneous trip faults in 220kV same-tower lines,’’ [37] Z. Zheng, Y. Yan, and J. Cai, ‘‘Non-specific TVLA method based on
Electr. Porcelain Surge Arresters, no. 4, pp. 65–71, 2018. double-sample KS test,’’ J. Commun., vol. 44, no. 5, pp. 137–147, 2023.
[14] H. Deng, S. Li, and D. Dai, ‘‘Risk assessment of lightning strikes [38] DSP Tian. Minimum Enclosing Rectangle. Accessed: May, 8, 2021.
on overhead transmission lines based on hierarchical and differentiated [Online]. Available: https://www.cnblogs.com/tiandsp/p/4044282.html
approach,’’ Power Syst. Protection Control, vol. 44, no. 4, pp. 69–75, 2016. [39] A. Yadav and D. Yadav, ‘‘Wavelet tree based hybrid geo-textual indexing
[15] L. Fang, S. Fang, and X. Xiong, ‘‘Probability analysis of indirect technique for geographical search,’’ Indian J. Sci. Technol., vol. 8, no. 33,
lightning faults in transmission lines and lightning forecasting methods,’’ pp. 624–632, Dec. 2015.
Guangdong Electr. Power, vol. 27, no. 3, pp. 95–100, 2014.
[16] Y. Liu, Y. Tian, and J. Xiong, ‘‘Research on lightning forecasting methods
for Guangzhou power grid,’’ Shaanxi Electr. Power, vol. 43, no. 7,
pp. 88–91, 2015. HONG HUO is currently the Leader of the
[17] B. Zhang, ‘‘Real-time prediction of lightning-induced trip probability for Transmission and Maintenance Innovation Work-
transmission lines based on lightning positioning system,’’ Ph.D. disser- shop, State Grid Dandong Electric Power Supply
tation, Dept. Power Syst. Automat., Southwest Jiaotong Univ., Chengdu, Company, and a Grid Craftsman with State Grid
China, 2014. Liaoning Electric Power Company Ltd.
[18] Z. Zhao, J. Shi, and Q. Yuan, ‘‘A new method for predicting lightning trip
probability of transmission lines,’’ Autom. Electric Power Syst., vol. 39,
no. 3, pp. 51–58, 2015.
[19] W. Wang, ‘‘Research on warning methods for wind-induced discharge and
lightning trip risk of transmission lines under severe convective weather,’’
Ph.D. dissertation, Dept. Elect. Eng., Chongqing Univ., Chongqing, China,
2017.
[20] S. Liu, X. Yang, and J. Yang, ‘‘Research on solutions to transmission DEZHI WANG received the degree from
capacity limitations of transmission corridors,’’ Northeast Electr. Power Shenyang Engineering University. He has been
Technol., vol. 38, no. 5, pp. 51–54, 2017. engaged in the maintenance and inspection of
[21] H. Liu, ‘‘Research on distribution line lightning protection scheme
transmission lines.
evaluation system based on improved lightning trip rate calculation
method,’’ Ph.D. dissertation, South China Univ. Technol., Guangzhou,
China, 2020.
[22] Y. Zhuang, Y. Xiao, and X. Chen, ‘‘Application of spatio-temporal
trajectory clustering algorithm in lightning risk assessment,’’ Guangdong
Meteorol., vol. 45, no. 3, pp. 95–98, 2023.
[23] M. Hahsler, M. Piekenbrock, and D. Doran, ‘‘DBSCAN: Fast density-
based clustering with R,’’ J. Stat. Softw., vol. 91, no. 1, pp. 1–30, 2019.
[24] N. Gholizadeh, H. Saadatfar, and N. Hanafi, ‘‘K-DBSCAN: An improved
HAO CHEN has been involved in transmission
DBSCAN algorithm for big data,’’ J. Supercomput., vol. 77, no. 6,
line maintenance work for many years, accumu-
pp. 6214–6235, Jun. 2021.
[25] P. Gao, H. Tian, and J. Li, ‘‘Thunderstorm mining and research based lating extensive on-site experience.
on improved DBSCAN algorithm,’’ High Voltage App., vol. 55, no. 4,
pp. 169–177, 2019.
[26] C. Jiang, ‘‘Research on spatio-temporal trajectory clustering of mobile
objects considering multidimensional semantics,’’ Ph.D. dissertation,
Dept. Geomatics Sci. Technol., Univ. Electron. Sci. Technol. China,
Chengdu, China, 2022.
[27] H. Liu, ‘‘Analysis of thunderstorm characteristics in South China
and potential forecast of XGBoost algorithm under multiple factors,’’
Ph.D. dissertation, Dept. Geomatics Eng. GNSS Meteorol., Guilin Univ.
Technol., Guangxi, China, 2021. CHUNLEI ZHAO has been consistently engaged
[28] S. Du, X. Liu, and H. Sun, ‘‘Study on the relationship between in the maintenance and inspection of transmission
dual-polarization radar parameters and lightning activity in a typical lines, accumulating rich on-site work experience.
thunderstorm process in South China,’’ J. Tropical Meteorol., vol. 37, no. 3,
pp. 427–438, 2021.
[29] W. Yan, C. Xiao, and X. Wu, ‘‘Calculation of lightning trip rate for 10 kV
distribution lines based on lightning activity characteristics,’’ High Voltage
Eng., vol. 47, no. 3, pp. 1118–1127, 2021.
[30] W. Qin, ‘‘Analysis of severe convective weather processes in Guangdong
province and research on lightning nowcasting warning,’’ Ph.D. disserta-
tion, Dept. Lightning Sci. Technol., Lightning Detection Warning, Nanjing
Univ. Inf. Sci., Nanjing, China, 2017.
[31] S. Ramraj, N. Uzir, R. Sunil, and S. Banerjee, ‘‘Experimenting XGBoost QI CHENG has been involved in the management
algorithm for prediction and classification of different datasets,’’ Int. of maintenance and inspection of transmission
J. Control Theory Appl., vol. 9, no. 40, pp. 651–662, 2016.
lines, providing guidance and participating in
[32] S. Dhaliwal, A.-A. Nahid, and R. Abbas, ‘‘Effective intrusion detection
system using XGBoost,’’ Information, vol. 9, no. 7, p. 149, Jun. 2018. numerous national and provincial-level transmis-
[33] L. Zhang, L. Kong, and H. Chen, ‘‘Hierarchical clustering method based on sion projects.
distance correlation coefficient,’’ Mathematica Numerica Sinica, vol. 41,
no. 3, pp. 320–334, 2019.
[34] C. Shen, S. Panda, and J. T. Vogelstein, ‘‘The chi-square test of distance
correlation,’’ J. Comput. Graph. Statist., vol. 31, no. 1, pp. 254–262,
Jan. 2022.

VOLUME 12, 2024 54181