基于改进卷积神经网络的电力工程数字化校核技术研究  

Research on digital checking technology of power engineering based on improved Convolutional Neural Network

作　　者：周鑫 周云浩 王楠[1] 李昊 韩志超 ZHOU Xin;ZHOU Yunhao;WANG Nan;LI Hao;HAN Zhichao(Electric Power Construction Engineering Consulting Branch,State Grid Beijing Electric Power Company,Beijing 100021,China)

机构地区：[1]国网北京市电力公司电力建设工程咨询分公司,北京100021

出　　处：《电子设计工程》2024年第9期147-151,共5页Electronic Design Engineering

基　　金：北京电力公司输变电工程应用项目(SGBJJS00XSJS2100639)。

摘　　要：针对传统电力工程验收过程使用人工费时费力且数据质量较差的问题,文中基于改进的卷积神经网络提出了一种电力工程数字化验收校核技术。该技术将Faster R-CNN作为基础模型,从3个方面对Faster R-CNN进行改进。使用ResNet网络代替原始基础网络,提升了算法局部特征的提取能力与运算效率。同时将K-means聚类算法与区域候选网络相结合,增强了模型的目标识别能力。再引入深度自编码网络作为预测网络,进而提高了算法的预测能力。在实验测试中,所提算法相较原始算法的准确率、召回率分别提升了3.7%和7.2%,可以对电力工程关键部件进行准确识别,有效节约了验收过程中的时间及人力成本。Aiming at the defects of traditional power engineering acceptance process,such as labor-consuming,laborious and poor data quality,a digital acceptance check technology for power engineering based on improved Convolutional Neural Network is proposed in this paper.The algorithm uses Fast R-CNN as the basic model and improves Fast R-CNN from three aspects.The ResNet network is used to replace the original basic network,which improves the local feature extraction ability and operation efficiency of the algorithm.Combining the K-means clustering algorithm with the regional candidate network,the target recognition ability of the model is enhanced.By introducing the deep self coding network as the prediction network,the prediction ability of the algorithm is effectively improved.In the experimental test,the accuracy and recall of the proposed algorithm are improved by 3.7%and 7.2%respectively compared with the original algorithm.It can also accurately identify key components of power engineering,effectively saving time and manpower costs in the acceptance process.

关 键 词：卷积神经网络 残差网络 K-MEANS聚类 深度自编码器 电力工程验收 

分 类 号：TN-9[电子电信]