摘要
电网设备的故障率分析是电网风险预警和运行评估的基础和关键,而电网设备多种故障因素与综合故障率的非线性权重关系的评估决定了故障率分析的准确度,为充分利用历史故障信息数据以准确评估多故障因素与综合故障率的权重关系,提出一种基于极限学习机(ELM)和迁移学习的电网设备多因素综合故障率分析方法;首先,设计了多源历史故障信息统计体系,构建电网设备故障分析样本数据库;将自身、过载、恶劣天气等作为主要故障因素,通过机器学习ELM算法,分析故障因素与综合故障率的权重关系,并利用多个强相关样本集的有效迁移提高机器学习的准确度,通过ELM结合迁移学习算法,克服了电网设备综合故障率分析中样本数量少且统计周期长的难题,充分利用历史故障信息,实现多因素设备综合故障率的准确评估,为风险评估和检修安排提高可靠的数据支撑;最后通过实际电网运行实例,验证了所提方法的可靠性和准确性。
The fault rate analysis of power grid equipment is the basis and key on risk assessment and operation assessment of power grid,the evaluation of nonlinear weight relationship with multiple failure factors and comprehensive failure rate of power grid equipment determines the accuracy of failure rate analysis,In order to make full use of historical fault factors and accurately assess the synergistic relationship of multiple failure factors and comprehensive failure rate,a multifactor comprehensive failure rate analysis method based on extreme learning machine(ELM)is proposed.Firstly,a multi-source historical fault information statistics system is designed,the sample database for fault analysis of power equipment is constructed.The relationship between the comprehensive fault rate and fault factors,such as equipment itself,overload,bad weather,is analyzed by ELM.The effective migration of multiple sample sets is proposed to improve the accuracy of machine learning.The analysis and calculation of comprehensive failure rate of equipment based on historical fault information can be realized.The algorithm based on ELM and transfer learning overcomes the difficult problem of small sample quantity and long statistical period in comprehensive failure rate analysis for power grid equipment.And it can improve the reliable data support for risk assessment and maintenance arrangement.Finally,the reliability and accuracy of the method are verified by the actual power grid operation data.
作者
陈丽惠
李哲
周键宇
马立熠
杨梓萌
CHEN Lihui;LI Zhe;ZHOU Jianyu;MA Liyi;YANG Zimeng(Baoshan Power Supply Bureau of Yunnan Power Grid Co.,Ltd.,Baoshan 678000,China)
出处
《计算机测量与控制》
2023年第4期30-35,48,共7页
Computer Measurement &Control
基金
南方电网科技项目(051200HA42210001)。
关键词
故障率
故障信息
极限学习机
多因素
迁移学习
fault rate
fault information
extreme learning machine
multivariate
transfer learning
