摘要
为解决利用机器学习算法在线生成电网运行断面时所面临的特征因素“维数灾”问题,提出了一种基于两层模式的电网运行断面特征选择与在线生成方法。上层为过滤式特征选择层,采用Fisher分和信息增益两种特征选择指标对初始因素集进行筛选,重点剔除重复因素和无关因素,输出基础因素集。下层为包裹式特征选择层,利用序列后向搜索算法,进一步分析电网运行断面与运行参数之间的内在关系,生成特征因素集,同步形成基于该特征因素集的运行断面生成智能体。基于某地区电网实际数据构造的算例表明,本文方法能大幅降低特征因素“维度”,与初始因素集相比缩小90%以上,基于该特征因素集的智能体能在10 s中内在线生成运行断面,准确性评价指标达到95%,能够满足电网实时运行控制辅助决策的需要。算例表明该方法能有效筛选运行断面辨识所需要的特征因素,对提升运行断面辨识准确率具有促进作用。
To solve the"curse of dimensionality"problem of feature factors in online generation of power network operation in machine learning algorithm,a two-tier structured method for feature selection of power system operation section and online generation was proposed.The upper tier was for filter feature selection,in which fisher score and information gain were used to screen the initial factor set.The repetitive and irrelevant factors were eliminated and the basic factor set was output.The lower tier was for envelope feature selection.The sequential backward search algorithm was used to further analyze the internal relationship between the operation section and the operation parameters of the power grid,to generate the feature factor set,and to synchronously form the operation section generation agent based on the feature factor set.A case study using actual data in a certain region power grid showes that the method can greatly reduce the"dimension"of characteristic factors and reduce it by more than 90%compared with the initial factor set.The intelligent energy based on the feature factor set can generate the running section of the inner line within 10 s,and the accuracy evaluation index reaches more than 90%,which well meet the needs of auxiliary decision-making for real-time operation control of power grid.In addition,numerical simulation shows that this method can effectively screen the characteristic factors needed for the identification of operating section and promote the accuracy of the identification of operating section.
作者
吴云亮
邓韦斯
姚海成
苏寅生
周毓敏
WU Yun-liang;DENG Wei-si;YAO Hai-cheng;SU Yin-sheng;ZHOU Yu-min(China Southern Power Grid Power Dispatch and Control Center,Guangzhou 510000,China)
出处
《科学技术与工程》
北大核心
2020年第27期11137-11142,共6页
Science Technology and Engineering
基金
广东省自然科学基金(2018A0303130134)
南方电网公司科技项目(0000002019030101XT00035)。
关键词
电网运行断面
在线生成
特征选择
机器学习
power system operation section
online generation
feature selection
machine learning
作者简介
第一作者:吴云亮(1984-),男,汉族,湖北十堰人,博士,高级工程师。研究方向:电力系统运行控制。E-mail:wuyunliang@csg.cn。
