基于数据融合并行特征提取的调峰电源设备状态评估方法  

Condition Assessment of Peaking Power Source Based on Data Fusion and Parallel Feature Extraction

作　　者：郝峰 方冰 祁炜雯 郭勤慧 朱传古 潘伟峰 HAO Feng;FANG Bing;QI Wei-wen;GUO Qin-hui;ZHU Chuan-gu;PAN Wei-feng(Water and Innovation Department,State Grid Corporation of China,Beijing 100031,China;Shaoxing Power Supply Company of State Grid Zhejiang Electric Power Co.,Ltd.,Shaoxing 310000,China;State Grid Electric Power Research Institute,Nanjing 211106,China)

机构地区：[1]国网水新部,北京100031 [2]国网浙江省电力有限公司绍兴供电公司,浙江绍兴310000 [3]国网电力科学研究院有限公司,江苏南京211106

出　　处：《水电能源科学》2023年第5期203-206,215,共5页Water Resources and Power

基　　金：国家电网有限公司总部科技项目(5700-202140381A-0-0-00)。

摘　　要：调峰电源设备状态评估有利于电力系统安全稳定运行。系统监测数据本质为一系列指标量的时间序列,评价系统状态时需综合考虑不同指标量间的耦合关系,且实时监测系统对模型的处理速度也有所要求。对此,提出了一种基于数据融合并行特征提取的调峰电源设备状态评估方法。并行提取时间序列统计特征,基于多重假设检验选择重要特征,再利用所设计的分层图卷积网络实现信息整合。试验表明,对比现有模型,该方法的识别准确率大幅提高,运行时间缩短,可迁移性提升。For the task of condition evaluation of peaking power equipment,the monitoring indicators are defined as a set of time series,among which the complex coupling relationships need to be considered.Besides,requirements on processing time are introduced by the real-time systems.A new framework is proposed to address these problems via the technology of parallel feature extraction as well as data fusion.The time series are statistically analyzed in parallel where multiple hypothesis testing is used to select the important features.Using a defined hierarchical graph convolutional network,the related information is integrated for the final condition evaluation task.Compared with the existing models,experiments indicate that the proposed method with stronger transferability and shorter processing time has a much higher predication accuracy.

关 键 词：调峰电源设备状态评估 并行时间序列处理 图网络 特征提取 数据融合 

分 类 号：TM712[电气工程—电力系统及自动化]