Early Warning Model of Power Grid Meteorological Disaster Based

on Machine Learning

Wei Chen, Xudong Wang, Jie Duan, Zhengyi Liu and Yibin Zhao

Shanxi Yitong Power Grid Protection Automation Co. Ltd, 0300021, China

Keywords: Machine Learning, Algorithm, Early Warning of Meteorological Disasters in Power Grids, Early Warning

Model.

Abstract: In order to solve the problem of early warning of meteorological disasters in power grid, this paper uses

random forest model modeling to construct an efficient early warning system based on the analysis of the

correlation between meteorological factors and power grid faults. In the process of research, this paper collects

the power grid operation data and meteorological records of a province for 5 years, and uses a data-driven

method to train and optimize the model. The experimental data show that the prediction accuracy of the model

is 88.7%, which is significantly better than the traditional method and has a strong application prospect. The

research results show that the early warning model of power grid meteorological disasters based on machine

learning can effectively improve the early warning ability of the power grid under complex meteorological

conditions after being constructed by using the random forest algorithm, and provide stable and strong support

for the safe operation of the power grid.

1 INTRODUCTION

Power grid meteorological disaster early warning is

one of the research directions to make the stable

operation of the power grid, due to the intensification

of climate change, so extreme weather is frequent,

which is easy to cause threats to the power system.

Previously, methods based on statistical analysis were

proposed to predict the impact of meteorological

disasters on power grids, but these methods were

unable to cope with complex meteorological

conditions due to the strong dependence and

inflexibility of data. Some researchers have

emphasized that this can be handled by physical

modeling, but it cannot be applied on a large scale

because it requires a large number of practical

parameters. The reason why this paper uses the

random forest model to carry out the research on

power grid meteorological disaster early warning is

that, as a machine learning algorithm, the random

forest algorithm can process high-dimensional data,

and in addition, it has a strong generalization ability,

which can effectively reduce false alarms and false

negatives in power grid meteorological disaster early

warning. It is hoped that the research in this paper can

effectively improve the disaster prevention ability of

the power grid system and ensure the stability and

security of power supply.

2 RELATED WORKS

2.1 Machine Learning Theory

Machine learning theory is one of the core theories of

the power grid meteorological disaster early warning

model in this paper. Machine learning is based on

building mathematical models, learning patterns from

data (Chen, Huang et al. 2023), and does not rely on

explicit programming instructions. Based on the

ensemble learning theory, the random forest

algorithm can be effectively used in the early warning

of meteorological disasters in power grids, and the

results can be effectively combined through the

training of multiple decision trees, so as to improve

its overall prediction performance (Chen and

Srivastava, 2023). Random forests can reduce

variance and prevent overfitting, which is suitable for

fault prediction under complex meteorological

conditions of power grids.

Chen, W., Wang, X., Duan, J., Liu, Z. and Zhao, Y.

Early Warning Model of Power Grid Meteorological Disaster Based on Machine Learning.

DOI: 10.5220/0013534600004664

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 18-24

ISBN: 978-989-758-763-4

2.2 Theory of Meteorological Disasters

in Power Grids

The theory of power grid meteorological disasters is

the basic theory for the study of the influence of

meteorological conditions on the operation of power

grid. The operation of grid equipment can be affected

by various meteorological factors, such as wind speed

and precipitation, temperature changes, etc. (Hu, Qu

et al. 2024). For example, strong winds can cause

power lines to break or collapse, and ice, snow and

freezing rain can cause transmission lines to freeze

and cause equipment failure. The theory of power

grid meteorological disasters is based on the analysis

of power grid failure modes under different

meteorological conditions (Ling, Chen, et al. 2023),

which provides guidance for the input feature

selection and data preparation of machine learning

models.

2.3 Data-Driven Modeling Theory

Data-driven modeling theory emphasizes that big

data and advanced analytics techniques can be used

to perform model construction and provide decision

support. In the early warning of power grid

meteorological disasters, data-driven modeling relies

on historical meteorological data and power grid fault

records, and based on the analysis of these data, a

correlation model between meteorological factors

and power grid faults can be established. This

approach does not rely too heavily on traditional

physical modeling, but uses a combination of data

pattern discovery and machine learning algorithms

(Liu and Chen, 2023) to improve the accuracy and

timeliness of its early warnings. For the early warning

of power grid meteorological disasters, the Random

Forest algorithm is used to construct the model. The

model is based on the integration of multiple decision

trees to effectively predict the risk of grid failure

under different meteorological conditions

(McLoughlin, Gifkins et al. 2023). The power grid

system is affected by a variety of meteorological

factors, such as strong winds and heavy rains, drastic

temperature changes, etc., which may cause potential

risks to power grid lines, substations, and power

equipment. Therefore, the early warning accuracy of

the model should be improved based on reasonable

parameter setting and optimization methods (Pandey,

and Basnet, 2023).

3 METHODS

3.1 Design of Meteorological Disaster

Early Warning Model

In the early warning of power grid meteorological

disasters, the setting of the number of decision trees

directly affects the robustness and performance of the

model. The random forest model is based on the

integration of multiple decision trees to reduce the

risk of overfitting a single tree. If the number of

decision trees is too small, the model may not be able

to fully capture its complex meteorological

characteristics, which may cause false alarms or false

negatives. If the number of trees is too large, the

computation cost will increase, which in turn will

affect the real-time performance. For this, see Eq. (1).

𝑦 =





∑

𝑇







(𝑋)

)

In equation (1), 𝑦is represents the final prediction

value of the power grid meteorological disaster

warning, which𝑛is refers to the number of decision

trees. 𝑇



(𝑋) is the prediction made by the ith decision

tree on various input characteristics X, including

wind speed or precipitation, temperature, etc. By

increasing the number of decision trees, the model

can effectively capture the relationship between

meteorological variables and power grid fault risk,

and improve the accuracy of early warning.

The maximum depth of the tree controls the

complexity of each decision tree. In the early warning

of power grid meteorological disasters, the shallow

tree depth may lead to the underfitting of the model

and the inability to capture the complex

meteorological disaster model. Too deep tree depth

may lead to overfitting of the model, which will be

too adaptable to historical meteorological data, and

perform poorly in the face of new meteorological

conditions (Wang, Wen, et al. 2023). Therefore, when

constructing the power grid meteorological disaster

early warning model, the depth of the tree should be

reasonably set according to the complexity of the

data. For this, see Eq. (2).

𝑇



(𝑋)=

∑

𝜔









ℎ



(𝑋)

)

In this formula, 𝑇



(𝑋) is the first tree to predict𝑖is

the power grid fault under specific meteorological

conditions, and ℎ



(𝑋) is the division planning of the

first node in𝑖is the tree. It𝑚 is the depth of the tree.

After setting the depth of the tree reasonably, the

model can better balance the different impacts of

Early Warning Model of Power Grid Meteorological Disaster Based on Machine Learning

complex meteorological conditions on power grid

equipment.

In the power grid meteorological disaster early

warning model, only a part of the meteorological

features are selected from the meteorological features

each time the decision tree nodes are divided to

perform the division. For example, a random

selection of meteorological variables such as wind

speed and temperature, humidity, and precipitation

prevents a specific meteorological variable, such as

wind speed, from dominating model decisions. This

random selection will enhance the diversity of the

model and thus avoid early warning errors due to

fluctuations in a variable. In this regard, Eq. (3) can

be seen.

𝐺(𝑋)=

∑

𝜔









𝑓



(𝑋)

(3)

In Eq. (3), 𝐺(𝑋) is the final predicted value, the

𝑓



(𝑋) is contribution of each selected meteorological

feature to its model decision. In the early warning of

power grid meteorological disasters, the random

combination of factors such as strong winds, heavy

rains, and drastic temperature changes, their

contribution to the final prediction of power grid

faults by the model needs to be stratified, which can

ensure that the stochastic algorithm model can make

accurate early warnings for a variety of

meteorological disasters.

3.2 Training and optimization of Power

Meteorological Disaster Early

Warning Model

During the model training process, the random forest

uses bootstrap sampling to generate multiple sub-

datasets. Specifically, each decision tree performs

training on a different subset. The advantage is that

the generalization ability of the model can be

improved, especially in the power grid

meteorological disasters, where the complexity of

meteorological factors is high, and each decision tree

needs to be trained based on a different subset of data,

under which the potential risks to the power grid

operation under different meteorological conditions

can be better captured (Zhang, and Song, 2023).

After the training is completed, the model can

effectively predict the impact of future

meteorological disasters on the power grid, such as

the probability of power grid failure under extreme

weather in a certain area.

In the power grid meteorological disaster early

warning model, the hyperparameters of the model

need to be tuned, which is also part of the

optimization process. For example, the number and

maximum depth of decision trees need to be tuned.

The tuning process can be based on grid search or

random search, and the optimal combination of

parameters can be found based on the minimization

of the loss function on the validation set. See Eq. (4)

for details.

𝜃

∗

= 𝑎𝑟𝑔𝑚𝑖𝑛



𝐿

val

(𝜃) （

)

In this equation,

（

𝜃

∗

）

is the optimal

combination of parameters is represented, 𝐿

val

(𝜃) is

refers to the loss function on its validation set.

Through the method of tuning these parameters, the

efficiency and accuracy of the model in the early

warning of power grid meteorological disasters will

be ensured, especially in the face of extreme weather

conditions. To effectively prevent the model from

overfitting on specific meteorological data, it would

be very effective to introduce regularization. In the

case of power grid meteorological disasters, some

extreme meteorological conditions may occur less

frequently in historical data, which can easily make

the model have problems or perform poorly in the

face of these conditions. Based on regularization, it

will be possible to effectively constrain the

complexity of the model and thus avoid overfitting.

For this, see Eq. (5).

𝐿



= 𝐿(𝑦, 𝑦)+𝜆

∑

𝑤







)

In equation (5), 𝐿

reg

is represents the loss function

after adding regularization, but 𝜆is the regularization

intensity, which can effectively improve the

generalization ability of the model in the early

warning of power grid meteorological disasters,

especially in the face of rare extreme weather.

In order to further improve the robustness of its

power grid meteorological disaster early warning

model, random forests can be integrated with other

model executions, such as gradient boosting decision

trees. This ensemble strategy can improve the

prediction accuracy of the model under complex

meteorological conditions based on the advantages of

integrating multiple algorithms. See Eq. (6) for this.

𝑦 = 𝛼 × 𝑦



+(1−𝛼)×𝑦



)

In Eq. (6), 𝑦



is the random forest prediction

value, 𝑦



is the prediction value of the gradient

boosting tree, and𝛼 is the weight parameter. Based on

the ensemble learning method, the prediction ability

of the power grid meteorological disaster early

warning model under a variety of meteorological

disasters will be effectively improved. The

INCOFT 2025 - International Conference on Futuristic Technology

application of the power grid meteorological disaster

early warning model based on random forest can

effectively capture the complex relationship between

meteorological factors and power grid faults. The

model uses reasonable parameter setting and

optimization methods to ensure the accuracy and

robustness of the early warning system, and provides

the best guarantee for the safe operation of the power

grid system.

4 RESULTS AND DISCUSSION

4.1 Case introduction of

Meteorological Disaster Early

Warning Model

The operation data and meteorological disaster

records of a provincial power grid in the past five

years are a case study in this paper. The region's

power grid covers a variety of terrain and climate

zones, and the meteorological conditions are complex

and changeable, often affected by various extreme

weather such as storms, thunder, lightning, and

freezing. These meteorological disasters frequently

cause damage to power equipment, interruption of

transmission lines, and disruption of power supply.

Therefore, the provincial power grid company hopes

to identify some potential risks in advance based on

the meteorological disaster early warning model

constructed this time, and then improve the disaster

prevention and resilience of the provincial power

grid, and then ensure the safety and stability of its

power supply, the warning range is shown in Figure

Figure 1 Scope of meteorological disaster warning for

power grid.

Undoubtedly, the main focus is on predicting the

power grid of coastal cities in areas with frequent

meteorological occurrences, and tracking and

analyzing meteorological disasters in the region.

4.2 The Overall Effectiveness of

Disaster Warning

In the research and development process of the power

grid meteorological disaster early warning model, in

order to verify the effectiveness and stability of the

model, it is necessary to carry out multiple rounds of

simulation experiments. The purpose of the

simulation is to test the iterative performance of the

model in a simulated environment, evaluate the

predictive ability of the model based on a large

number of historical meteorological data and grid

fault data (Zhang, Xie et al. 2023), and observe its

performance under different parameter

configurations. During the simulation process, the

model gradually improves its prediction accuracy

based on continuous optimization, and finally lays the

foundation for the next practical application

verification. Table I shows the iterative results of the

model during the simulation process, including the

training set, the validation set, the accuracy of the test

set, and the time required for each iteration. After 60

iterations, it can be seen that the prediction

performance of the model has gradually improved

and tends to be stable. The simulation test iteration

was 60 times, 600 verifications, 40 days of testing,

200 people participated, referring to 5 years of

meteorological data, 5 types of meteorological

disasters.

Table 1: Results of the simulation of this model

The

number of

iterations

Training

set

accurac

Validation

set

accurac

Test set

accuracy

Time per

iteration

(seconds)

Wind

disasters.

0.754 0.733 0.7225 5.2

Tsunami

disaster.

0.758 0.736 0.7250 5.4

Earthquake

disasters.

0.762 0.739 0.7275 5.6

Random

disasters.

0.766 0.742 0.7300 5.8

Sudden

disasters.

0.770 0.745 0.7325 6.0

... ... ... ... ...

Overall

disaster

situation.

0.905 0.887 0.8730 9.2

Early Warning Model of Power Grid Meteorological Disaster Based on Machine Learning

Based on the simulation process, it can be seen:

Its accuracy is significantly improved. Under the

continuous iteration of the model, its accuracy on the

training set, verification set and test set is gradually

improved, which shows that the power grid

meteorological disaster early warning model can

effectively capture the complex relationship between

meteorological conditions and power grid faults

through the random forest algorithm, one of the

machine learning methods,The degree of

meteorological disaster prediction within the green

range is shown in Figure 2.

Figure 2: Predicted degree of meteorological disasters in the

power grid.

As analyzed in Figure 2, it can be seen that the

prediction level of power grid component disasters is

relatively high, but its prediction range needs to be

further determined. The specific results are shown in

Figure 3.

Figure 3: The predicted range of meteorological disasters.

The measurement range of meteorological

disaster prediction was found to be relatively good.

4.3 Prediction and Effectiveness

Assessment of Meteorological

Disasters in the Power Grid

Its stability is significantly enhanced, which indicates

that the performance indicators of the model tend to

be stable during the simulation process. In other

words, it is proved that the model has a strong

generalization ability after a period of optimization.

Moreover, the time consumption is moderate, and the

time of each iteration increases under the increasing

complexity of the model, but it is still within the

acceptable range, so as to meet the real-time

requirements of power grid early warning, as shown

in Table 2.

Table 2: Grid operation data and meteorological disaster

record data of the province in the past five years.

year Number of

meteorolo

gical

disasters

The

number of

power

equipment

failures

The number

transmission

line

interruptions

Hours of

power

supply

interruption

2018 15 10 8 12

2019 20 14 10 18

2020 18 13 9 16

2021 22 16 11 20

2022 19 12 10 15

The table shows the number of meteorological

disasters in the province's power grid over the past

five years, the corresponding power equipment

failures, transmission line interruptions, and the

number of hours of power supply interruptions. It can

be seen that the annual meteorological disasters have

different degrees of impact on the operation of the

power grid in the province, which provides rich

historical data for the application of the model, The

results are shown in Table 3.

Table 3: The failure risk identified by the model and the

actual situation.

Meteorologi

cal type

Wind

speed

(m/s)

Precip

itation

(mm)

mpe

ratu

(°C)

Risks of

failures

identified

(number

of times)

The actual

number of

failures

that

occurre

Storm 25 150 18 17 16

Thunderstor

18 55 23 12 12

Freezing

rain

7 30 -3 22 20

High

tem

erature

0 0 36 9 8

Heavy rain 10 90 21 10 9

Table 3 shows the prediction results given by the grid

meteorological disaster early warning model in

identifying the risk of grid failure under different

meteorological conditions, and comparing it with the

INCOFT 2025 - International Conference on Futuristic Technology

actual fault situation. The table shows that the model

has a high degree of accuracy under extreme

meteorological conditions, especially in storms,

freezing rain and other weather, and its prediction

results are quite close to the actual failure situation.

4.4 Effect of Meteorological Disaster

Early Warning Model

The early warning effect is remarkable. This can be

seen in combination with the data. Specifically, the

accuracy of the model provides a very effective

decision-making basis for power grid management,

which can accurately and advance the risk caused by

extreme weather, which is conducive to the provincial

power grid to take early protective measures and

reduce power interruptions and equipment damage,

The overall change result is shown in Figure 4.

Figure 4: Overall prediction effect of meteorological

disasters.

This can be seen through a comprehensive

analysis of the key data from Table 2 and Table 3.

Storms and freezing rain are the meteorological

conditions that have the most significant impact on

the grid, and the gap between the model prediction

and the actual failure is very small. For example, in

storm conditions, there are 17 predicted failures and

16 actual occurrences. Freezing rain was predicted 22

times and actually occurred 20 times, indicating that

the model is effective in dealing with the risk of

extreme weather. The model still shows high

accuracy under relatively low-risk weather conditions

such as thunderstorms and high temperatures. There

were 12 predicted failures in thunderstorms, which

was consistent with the actual failures, and in hot

weather, there were 9 predictions, but the actual

number was 8.

5 CONCLUSIONS

This paper constructs a random forest-based power

grid meteorological disaster early warning model to

prove the effectiveness and applicability of the model

under complex meteorological conditions. Based on

the comprehensive analysis of the correlation

between meteorological factors and power grid faults,

the model can accurately identify potential power grid

risks and warn in advance of the possible impact of

various extreme meteorological events. Compared

with the traditional method, the proposed model has

stronger prediction ability, which can reduce false

positives and false negatives, and provide scientific

and effective decision support for power grid

managers. At the same time, the application of this

model can effectively improve the disaster prevention

and resilience of the power grid, and provide a strong

guarantee for the safe operation of the power system.

To a certain extent, the research in this paper has been

very complete, but its data content still has certain

limitations and needs to be expanded in the future.

REFERENCES

Chen, Z., G. W. Huang, W. Xie, Y. Z. Zhang, and L. Wang.

(2023). GNSS Real-Time Warning Technology for

Expansive Soil Landslide-A Case in Ningming

Demonstration Area. Remote Sensing, 15(11).

Chen, Z. H., and G. Srivastava. (2023). Security Threat

Early Warning of Distance Education System Based on

Blockchain. Journal of Internet Technology, 24(5),

1149-1157.

Hu, S. H., M. X. Qu, Y. C. Yuan, and Z. K. Pan. (2024).

Coupling cloud theory and concept hierarchy

construction early warning thresholds for deformation

safety of tailings dam. Natural Hazards, 120(9), 8827-

8849.

Ling, M. H., J. Y. Chen, P. Z. Zhang, X. L. Wei, and L. L.

Yu. (2023). Early warning methods on the carrying

capacity of regional groundwater resources. Water

Supply, 23(8), 3179-3191.

Liu, Q. X., and Z. W. Chen. (2023). Early warning control

model and simulation study of engineering safety risk

based on a convolutional neural network. Neural

Computing & Applications, 35(35), 24587-24594.

McLoughlin, S., J. Gifkins, and A. J. Bellamy. (2023). The

Evolution of Mass Atrocity Early Warning in the UN

Secretariat: Fit for Purpose? International

Peacekeeping, 30(4), 477-505.

Pandey, C. L., and A. Basnet. (2023). Challenges and

Prospects of Flood Early Warning Systems: A Study of

Narayani Basin. Asian Journal of Water Environment

and Pollution, 20(1), 17-24.

Wang, Y. B., J. H. Wen, W. Zhou, B. M. Tao, Q. W. Wu,

C. L. Fu, and H. Li. (2023). An early warning method

for abnormal behavior of college students based on

multimodal fusion and improved decision tree. Journal

of Intelligent & Fuzzy Systems, 45(5), 8405-8427.

Zhang, X. F., and C. X. Song. (2023). Study on safety early-

warning model of bridge underwater pile foundations.

Early Warning Model of Power Grid Meteorological Disaster Based on Machine Learning

Structural Monitoring and Maintenance, an

International Journal, 10(2), 107-116.

Zhang, Y. Q., H. Xie, and L. L. Yan. (2023). Analysis of

the application of Z3 model in the early warning of

corporate financial risk. Reviews of Adhesion and

Adhesives, 11(3), 612-630.

INCOFT 2025 - International Conference on Futuristic Technology