基于多列神经网络的用电行为自诊断技术研究  

Research on Power Consumption Self-diagnosis Technology Based on Multi-column Neural Network

作　　者：王鹏飞 汤铭 杜元翰 李效龙 WANG Pengfei;TANG Ming;DU Yuanhan;LI Xiaolong(Information&Telecommunication Branch of State Grid Jiangsu Electric Power Co.,Ltd.,Nanjing 210024,China;Information System Integration Branch of NARI Technology Co.,Ltd.,Nanjing 211100,China)

机构地区：[1]国网江苏省电力有限公司信息通信分公司,江苏南京210024 [2]国电南瑞科技股份有限公司信息系统集成分公司,江苏南京211100

出　　处：《微型电脑应用》2023年第1期111-113,共3页Microcomputer Applications

摘　　要：为了进一步提升电能计量的整体自诊断管理,使用3列相对独立的多列神经网络算法,在有限且不完备数据的基础上,利用全网计量节点提供的上行、下行尖峰平谷计量数据,实现对计量系统故障、用电器配置变更、用户窃电等三大主要可能性做出基于大数据深度挖掘的主动判断。在计量系统故障、用电系统变更、窃电行为影响和总和敏感性方面,该系统较早期方法分别实现了14.3%、83.7%、36.3%、34.8%的计算效能提升,对未来电力计量系统的整体管理效率提升有积极意义。This paper used 3-column relatively independent multi-column neural network to further improve the overall self-diagnosis management of power metering. Based on the limited and incomplete data, we used the upstream and downstream peak and valley metering data provided by the metering nodes of the whole network. The algorithm proposed in this paper realized the active judgment based on big data in-depth mining on the three main possibilities of metering system failure, electrical appliance configuration change and user power theft. In the aspects of metering system failure, power system change, influence of electricity stealing behavior and sum sensitivity, the calculation efficiency of the innovation system is improved by 14.3%, 83.7%, 36.3% and 34.8%, respectively. This is of positive significance to improve the overall management efficiency of power metering system in the future.

关 键 词：用电信息采集 自诊断 数据同构化 多列神经网络 数据升维算法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TM933.4[自动化与计算机技术—控制科学与工程]