基于深度学习的输电线路故障预测方法研究.pdfVIP

DOI:10.13882/ki.ncdqh.2024.02.001NewDistributionNetwork|新型配电网

基于深度学习的输电线路故障预测方法研究

茹洁宣

（北京博望华科科技有限公司工程咨询分公司，北京丰台100071）

摘要：

输电线路的故障会给电力系统带来重大影响，因此对其进行准确预测变得尤为重要，文章旨在探

索如何有效地结合现有输电线路监控系统与先进的深度学习技术，以实现对故障的预测。本研究选择了

基于MobileNet架构的卷积神经网络，采用了深度分离卷积、Transformer注意力机制、多尺度特征提取等

技术，训练过程中还采用了迁移学习和域自适应技术来增强模型的泛化能力。在验证数据集上，本研究

算法实现了更短的检测时间和更高的正确率，优于VGG16和原卷积神经网络，也证明了模型在各种场景

下的鲁棒性。本次探索为输电线路的维护和预测提供了可行的方法，也为电力行业在实际应用中集成深

度学习技术提供参考。

关键词：

深度学习；卷积神经网络；输电线路；故障预测中图分类号：

TM763

Exploration

of

Transmission

Line

Fault

Prediction

Methods

Based

on

Deep

Learning

RU

Jiexuan

(Beijing

Bowang

Huake

Science

and

Technology

Co.,

Ltd.,

Beijing

Fengtai

100071,

China)

Abstract:

Faults

in

transmission

lines

can

have

significant

impacts

on

the

power

system,

making

accurate

prediction

of

such

faults

particularly

crucial.

This

paper

aims

to

explore

how

to

effectively

integrate

existing

transmission

line

monitoring

systems

with

advanced

deep

learning

technologies

to

develop

fault

prediction

methods.

This

research

adopted

a

convolutional

neural

network

based

on

the

MobileNet

architecture,

utilizing

techniques

such

as

depthwise

separable

convolutions,

Transformer

attention

mechanisms,

and

multi-scale

feature

extraction.

Additionally,

transfer

learning

and

domain

adaptation

techniques

were

employed

during

the

training

process

to

enhance

the

models

generalization

capabilities.

On

the

validation

dataset,

the

proposed

algorithm

achieved

shorter

detection

times

and

higher

accuracy

rates,

outperforming

both

VGG16

and

traditional

convolutional

neural

networks.

This

also

demonstrated

the

robustness

of

the

model

across

various

scenarios.

Keywords:

deep

learning;

convo