基于用户协同过滤的购电推荐算法  

Power purchase recommendation algorithm based on user collaborative filtering

作　　者：荣以平 张鹏 朱伟义 张敏 代佰华 张利[3] RONG Yiping;ZHANG Peng;ZHU Weiyi;ZHANG Min;DAI Baihua;ZHANG Li(Marketing Departmant of State Grid Shandong Electric Power Group Company,Jinan 250001,China;State Grid Shandong Binzhou Power Supply Company,Binzhou 256600,China;School of Electrical Engineering,Shandong University,Jinan 250061,China)

机构地区：[1]国网山东省电力集团公司营销部,济南250001 [2]国网山东滨州供电公司,山东滨州256600 [3]山东大学电气工程学院,济南250061

出　　处：《电力需求侧管理》2020年第5期58-62,共5页Power Demand Side Management

基　　金：国家电网公司科技项目(SGSDBZ00YXJS1800450)。

摘　　要：售电市场放开后,电力大用户面临如何从众多发电商中找到合适交易对象的难题,而电网公司作为服务型企业也有为购电用户提供推荐服务的良好意愿。针对这一新需求,提出了基于用户协同过滤的购电推荐算法。在无法获得反映用户购电行为偏好的显性历史数据的前提下,建立了用户行为的隐性反馈数据集;在分析了多种相似度计算方法的应用效果后,为消除数据集中各属性元素数值差异较大造成的相似度计算不合理性,提出了以马氏距离为核心度量以形成相似用户集的方法;进一步定义了表征用户对电厂购电偏好的购电量占比,以此作为进行购电交易对象推荐的排序依据。采用山东省电力市场交易数据进行了算例仿真,结果验证了算法的可行性。After the opening of the electricity retail market,large power users are faced with the problem of how to find a suitable trading partner from a large number of power producers.As a service-oriented enterprise,power grid companies have a good will to providerecommendation service for power consumers.In order to meet this new demand,a power purchase recommendation algorithm based on user collaborative filtering is proposed.In order to eliminate the unreasonable similarity calculation caused by the large difference of attribute elements in the data set,a new method based on Mahalanobis distance is proposed to form a similar user set.Furthermore,the proportion of purchasing power amount that represents the user’s preference for power plant is defined,which is used as the order basis for the recommendation of purchasing trading partners.Case studies based on Shandong electric power market transaction data are carried out.The results show that the algorithm is feasible.

关 键 词：大用户直购电 基于用户的协同过滤算法 马氏距离 购电推荐 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术] F426.61[自动化与计算机技术—计算机科学与技术] F274[经济管理—产业经济]