摘要
电力用户价值画像对于提升我国供电企业的营销服务资源配置效能,提高智慧营销管理水平,从而最大化撬动供电企业的综合效益具有重要的现实意义。现有工业电力用户价值画像模型评级指标体系未能反映国家最新能源政策对我国工业企业的发展要求、无法良好应对现实电力用户用电数据的稀疏性,以及对于新用户价值等级预测的准确性存在提高空间。针对以上问题,本文集成数据挖掘技术中的谱聚类算法(Spectral Clustering,SC)与智能算法中的粗糙集理论(Rough Set,RS),构建了一种优化的数据驱动型工业电力用户价值画像模型,简称SC-RS模型。新模型构造围绕“知识萃取-知识推理-知识服务”的逻辑脉络展开,首先,在“知识萃取”部分,结合我国“碳达峰”目标与“碳中和”愿景,构建优化的我国工业电力用户价值评级指标体系,此外,采用谱聚类技术,并联合网格搜索策略,提炼用户价值等级信息情报;然后,在“知识推理”部分,应用粗糙集理论,继承已有三维规则挖掘框架,构建基于ChiMerge离散法与变异系数的行约简、基于系统依赖度的列约简、基于对象确定性因子的格约简,以及基于规则强度的规则提取方案这一拓展的四维规则挖掘模型,生成用户价值等级规则库;最后,在“知识服务”部分,一方面运用用户价值等级信息情报,构造价值决策系统,以及描摹群体用户价值画像,另一方面运用规则库,呈现可理解的价值知识,以及构造价值等级规则软分类器,实现新用户价值等级预测与个体价值画像描摹。为了展示模型的应用路径与具体步骤,采用实际工业电力用户数据,对模型开展实证研究。结果表明,SC-RS模型构建的评级指标体系紧跟我国电力行业最新发展动态,具有较强先进性;模型能够兼容稀疏性数据,对数据要求低;构造的粗糙四维规则挖掘模型可行有效,且对新用户价值等级的预测准确性高。综上,SC-RS模型能够对电力用户大数据实现价值情报挖掘与利用,是为我国电力企业智慧营销管理赋能的有力工具。
For Chinese power supply enterprises, the value portrait of power customers has important practical significance for improving the resource allocation efficiency of marketing service, promoting the smart marketing management, and therefore maximizing the comprehensive benefits. However, the reported models fail to reflect the development requirements of China’s industrial enterprises in the latest national energy policy, cannot cope well with the sparsity of real power customers’ electricity consumption data, and has room for improvement in the accuracy of new customers’ value rating prediction. Therefore, based on the spectral clustering(SC) technology and the rough set(RS) theory, an optimized industrial power customer value portrait model termed as SC-RS is constructed. The new model is constructed according to the logical framework of “knowledge extraction-knowledge reasoning-knowledge service”. Firstly, in the knowledge extraction part, an optimized China’s industrial power customer rating index system is constructed, based on the China’s “Peak Carbon Dioxide Emissions” goal and “Carbon Neutrality” vision. In addition, the grid search strategy combined with the SC technology is used to refine user’s value grade information. Secondly, in the knowledge reasoning part, the RS theory is used to build a four-dimensional rule mining model to generate the rule base of the user value grade, based on the framework of three-dimensional rule mining. It consists of row reduction based on the ChiMerge discretization method and coefficient of variation, column reduction based on the system dependence, cell reduction based on the object certainty factor, and rule extraction based on the rule strength. Finally, in the knowledge service part, the user value grade intelligence is applied to portray the value portrait of group users. The rule base is used to present understandable value knowledge and construct a ruled-soft classifier to achieve value grade prediction and individual value portrait description of new users. The model is further applied on the data of actual industrial power customers. The results show that the constructed rating index system closely copes with the latest developments in Chinese power industry. The SC-RS model is compatible with sparse data and has low data requirements. The constructed four-dimensional rule mining model based on RS theory is feasible and effective, and can robustly predict the value grade of new customers. Moreover, the SC-RS model can realize the value intelligence mining and utilization of big data of electric power customers, which is a powerful tool for empowering the smart marketing management of China’s electric power enterprises.
作者
余顺坤
闫泓序
杜诗悦
林依青
YU Shun-kun;YAN Hong-xu;DU Shi-yue;LIN Yi-qing(School of Economics and Management,North China Electric Power University,Beijing 102206,China;Pumped-storage Technological Economic Research Institute,State Grid Xinyuan Co.,Ltd.,Beijing 100761,China;Shantou Power Supply Bureau of Guangdong Power Grid Co.,Ltd.,Shantou 515000,China)
出处
《中国管理科学》
CSSCI
CSCD
北大核心
2022年第3期106-116,共11页
Chinese Journal of Management Science
基金
中央高校基本科研业务费专项资金资助(2018QN066)。
关键词
智慧营销管理
电力用户价值画像模型
粗糙规则挖掘模型
价值等级预测
碳达峰
碳中和
smart marketing management
power customer value portrait model
rough rule mining model
value grade prediction
carbon peak
carbon neutral
