摘要
针对电力设备因素、人为因素与气象因素导致的电网故障问题,提出了一种基于自编码神经网络的电网故障预测算法.利用电网运维数据与气象数据来预训练自编码网络,以提取其特征及不同数据间的关联关系.基于多级别特征融合和预测网络来构建各影响因素与电网故障间的映射,根据当前电网环境得到电网故障类型,并使用稀疏正则项来增强网络的鲁棒性.仿真与算例分析结果表明,所提出的算法能够提取出丰富的气象特征,并准确预判电网在给定条件下是否会发生故障及发生故障的概率.
Aiming at the power grid faults caused by power equipment factors,human factors and meteorological factors,a fault prediction algorithm based on autoencoding neural network for power grid was proposed.Power grid operation and maintenance data as well as meteorological data were used to pre-train autoencoding networks for the extraction of their features and the associations among different data.According to multi-level feature fusion and prediction networks,mappings among various influencing factors and grid faults were built.In view of current power grid environment,grid fault types were obtained,and sparse regular terms were used to enhance the robustness of networks.Simulation and example analysis results show that the as-proposed algorithm can extract abundant meteorological features and accurately predict whether or not the power grid will fail as well as the fault probability under given conditions.
作者
严宇平
萧展辉
YAN Yu-ping;XIAO Zhan-hui(Digital Department,Guangdong Power Grid Co.Ltd.,Guangzhou 510000,China)
出处
《沈阳工业大学学报》
CAS
北大核心
2023年第1期1-5,共5页
Journal of Shenyang University of Technology
基金
国家重点研发计划项目(2017YFB0903504)。
关键词
电网
故障预测
自编码网络
多尺度特征
气象
数据处理
去噪
特征融合
power grid
fault prediction
autoencoding network
multi-scale feature
meteorology
data processing
denoising
feature fusion
