摘要
针对售电公司实现多样化服务类型,吸引更多用户的需求,提出了一种基于差异化特征提取的用户分层聚类方法,并对传统的k-medoids聚类算法进行改进,实现了聚类数目可变的自适应k-medoids算法。分层聚类中第1层聚类先基于马尔科夫模型提取代表用户用电行为多样性的用电特征,并运用自适应的k-medoids聚类算法实现对用户用电行为多变与否的识别。第2层聚类首先针对第1层聚类得到的各类用户提取差异化的用电特征,接着分别运用合适的聚类算法实现用户的再次分类。最后,为两层聚类后的子类用户推荐合适的电价套餐。实验结果表明,基于该差异化特征提取的分层聚类方法能够为售电公司实现有效的用户差异化套餐推荐服务,进而为吸引更多用户购电、扩大售电公司规模提供技术支撑。
To meet requirements of electricity retail companies providing variety of services and attracting more users, this paper proposed a user＇s hierarchical clustering method based on differentiated feature extraction and improved traditional k-medoids clustering algorithm to achieve adaptive k-medoids algorithm with variable numbers of clusters. The first-layer clustering used adaptive k-medoids algorithm to divide all users into several groups based on users＇ behavior and diversity of users＇ behavior was extracted with Markov model. The second-layer clustering extracted different features from different user groups obtained in the first-layer clustering. At the same time, different algorithms could be used to realize the second-layer clustering for different user groups. Finally, suitable price packages for sub-class users after two stages of clustering were recommended. Experimental results show that the user＇s hierarchical clustering method based on differentiated feature extraction can provide technical support for electricity retail companies to achieve effective pricing package recommendation services and attract more users to expand the scale of electricity companies.
出处
《电网技术》
EI
CSCD
北大核心
2018年第2期447-454,共8页
Power System Technology
基金
国家电网公司科技项目(智能电网用户行为理论与互动化模式研究)~~
关键词
售电侧改革
电价套餐推荐
用户分层聚类
差异化特征提取
自适应k-medoids算法
马尔科夫模型
electricity retail market reform
pricingpackage recommendation
users＇ hierarchical clusteringmethod
differentiated feature extraction
adaptive k-medoidsalgorithm
Markov model
