文章摘要
冷喜武,陈国平,蒋宇,等.智能电网监控运行大数据应用模型构建方法[J].电力系统自动化,2018,42(20):115-122. DOI: 10.7500/AEPS20170928003.
LENG Xiwu,CHEN Guoping,JIANG Yu, et al.Model Construction Method of Big Data Application for Monitoring and Control of Smart Grid[J].Automation of Electric Power Systems,2018,42(20):115-122. DOI: 10.7500/AEPS20170928003.
智能电网监控运行大数据应用模型构建方法
Model Construction Method of Big Data Application for Monitoring and Control of Smart Grid
DOI:10.7500/AEPS20170928003
关键词: 相关性分析  格兰杰因果关系  概率图模型  大数据应用建模
KeyWords: correlation analysis  Granger causal relationship  probabilistic graphical models  big data application modeling
上网日期:2018-08-13
基金项目:
作者单位E-mail
冷喜武 国家电网有限公司, 北京市 100031 xiwu-leng@sgcc.com.cn 
陈国平 国家电网有限公司, 北京市 100031  
蒋宇 国网江苏省电力有限公司, 江苏省南京市 210024  
张家琪 国网物资有限公司, 北京市 100120  
肖飞 国网上海市电力公司, 上海市 200122  
摘要:
      由于电网数据具有多源、高维、先验、异构的特点,并且蕴含大量因果关系,如何从电网历史大数据中挖掘出不同应用场景的因果关系,从而构建监控运行业务应用是电网监控专业实际生产及管理的热点问题。文中根据电网大数据的因果关系特征,提出了一种新颖的基于因果关联分析和概率图模型建模的智能挖掘算法框架。结合相关性分析和格兰杰因果分析方法,该框架能够从海量多源多维生数据中提炼出因果关系变量集,进而生成监控大数据因果概率图模型。基于上述理论,提出一种两阶段的电网监控运行大数据应用模型的工程构建方法,第1阶段采用大数据分析方法从海量多源多维生数据中挖掘出存在强因果关系的变量集合,第2阶段通过人工经验与算法推荐的因果概率图模型相结合,构造出监控业务模型。基于上述模型构建的智能电网监控运行大数据分析系统,已在多个省级调控中心上线运行,其运行结果验证了所提方法的有效性。
Abstract:
      The data from power grid not only have the feathers of multi-source, high-dimension, prior and hierarchical structure, but also contain enormous characteristics of causal relationship. It is the hotspot issue in practical manufacture and management of power grid monitoring system, which involves how to figure out causal relation models in different practical scenarios from historical big data of power grid, and to establish monitoring operation business. According to causal relation characteristics of big data for power grid, this paper proposes a novel intelligent data mining framework based on causal relative probabilistic graphical models. Through the combination of correlation analysis and Granger causal analysis, this framework can abstract correlative deductive mechanics of causal factor sets from enormous multi-sourced multi-dimensional raw data, and generate causal relative probabilistic graphical models of monitoring big data. After that, a construction method of application model for big data monitored by power grid with two stages is proposed. In the first phase, the big data analysis approaches are used to work out the variable set with strong causal relationship from mass multi-sourced and multi-dimensional data. In the second phase, the combination of expertise and recommended models is leveraged to filter out the set of functions so as to generate the business models of monitoring. The first big data real-time monitoring system of China, which is based on above methods, has already operated in five provincial control centers, and the final operating results validate the effectiveness of the proposed method.
查看全文(Free!)   查看附录   查看/发表评论  下载PDF阅读器