摘要
针对当前电力工程数据质量较差的问题,文中开展了基于深度学习的电力工程数据异常检测模型设计研究。提出了局部密度因子的改进方案,设计了一种基于深度自编码器(DAE)与高斯过程回归(GPR)的电力异常数据检测算法。该算法利用DAE模型实现了电力工程数据的重构,且将改进的局部密度因子、编码器输出数据及重构误差等作为GPR模型的输入,进而完成对异常数据的精准检测。仿真算例结果表明,与DAE、AE算法相比,所提算法的准确率可达89.2%,且稳定性更强。同时在实际应用中还可发现,通过加强对工程量及费用类型数据的校核管控,能够有效提升电力工程数据的质量,从而为电网的精细化运营提供数据基础。
In order to solve the problem of poor quality of current power engineering data,the design and research of power engineering data anomaly detection model based on deep learning is carried out in this paper.An improved scheme of local density factor is proposed,and a power anomaly data detection algorithm based on Deep Auto Encoder(DAE)and Gaussian Process Regression(GPR)is designed.The algorithm uses the DAE model to realize the data reconstruction of power engineering,and takes the improved local density factor,encoder output data and reconstruction error as the input of GPR model to realize the accurate detection of abnormal data.The simulation results show that the accuracy of the proposed algorithm can reach 89.2%and the stability is stronger than that of the DAE and AE algorithms.In practical application,it is found that the quality of power engineering data can be effectively improved and the data foundation for the refined operation of power grid can be provided by strengthening the verification and control of engineering quantity and cost type data.
作者
王斌
房向阳
毛华
孙岳
WANG Bin;FANG Xiangyang;MAO Hua;SUN Yue(State Grid Tianjin Electric Power Company,Tianjin 300010,China)
出处
《电子设计工程》
2024年第2期111-115,共5页
Electronic Design Engineering
基金
2021年国家电网公司基建部新技术研究项目(SGTJJS00JGJS2200092)。
关键词
深度学习
异常检测
高斯过程回归
深度自编码器
deep learning
anomaly detection
Gaussian Process Regression
deep auto encoder
