Research On The Industrial Electricity Customer Segmentation Based On Customer Value

Customer segmentation is the premise of enterprises to capture changing customer needs,target service objects,allocate service resources and carry out differentiated marketing or service strategies to achieve the consolidation and development of customer relationships.Traditional customer segmentation researches are usually for the fields of the retailing,financial industry and communications industry.Compared to these markets,the electricity market has some unique properties including: power has both characteristics of commodities and energy,and the relationship between customers and the power company is a contractual and exclusive relationship.Together with the diversity of customer categories,these features create challenges for customer segmentation in the power industry.As the power market of China is changing gradually from a vertical monopoly market to a market-oriented one and showing oversupply situation,the traditional rough-type classification manner has no longer met the requirements of today's advanced services.The power industry has an urgent need to implement deep segmentation under the guidance of customer value.In addition,the development of information engineering and the construction of smart grid,on the one hand,have reduce the difficulties of customer information obtaining and storing,on the other hand,the corresponding high-dimensional,uncertainty,mixed type and time-vary data of customers also proposed new requirements to exiting segmentation methods and techniques.In view of this,based on a comprehensive overview of the theory about customer segmentation and domestic and foreign researches of electricity customer segmentation,as the research object of the industrial electricity customer,issues of customer segmentation were studied from different sides like segmentation variables,segmentation methods and segmentation models,respectively.Firstly,a segmentation model which considering the needs of dynamic customer segmentation and including two stages of clustering process based on data mining was established.Then several key issues involved in this model were studied sequentially,which containing: elements of customer value of industrial electricity customers were analyzed under the guidance of the customer value theory and variables which measure these elements were also extracted.A dissimilarity measure model for the mixed attribute data was established to make up the shortcomings of traditional metrics in the dissimilarity measurement of categorical attributes with hierarchical structure.Then this dissertation goes deep into research on the adaptive evolutionary clustering technique for dynamic electricity customer segmentation.At last,a real segmentation instance of the industrial electricity customer is used to explain and validate the research results.The results and originative work of this paper are mainly reflected in the following aspects:(1)A segmentation model of the industrial electricity customer comprises a two-stage clustering process is proposed.After analyzing the basic theory of customer segmentation,the priori value model and the model based on data mining by comparing their advantages and disadvantages in modeling ideas,and the needs of dynamic customer segmentation,a segmentation model of the industrial electricity customer comprises a two-stage clustering process is presented by using the cross-industry standard data mining process(CRISP-DM)modeling framework.(2)Segmentation variables which indicate the value of the industrial electricity customer are extracted based on the analysis of the value constitution of this type of customer.Based on the analysis of the definition and connotation of customer value,the segmentation aims of industrial electricity customer are determined for electricity using features of such customers.On this basis,the value constitution of the industrial electricity customer is proposed which contains four dimensions namely the current value,the potential value,the energy efficiency contribution and the potentiality of sustainable development,respectively.Further more,under these dimensions,segmentation variables for the industrial electricity customer based on customer value are extracted by using the existing research outcomes,the demand side management theory and the energy service theory.According to the status of available data,the set of segmentation variables is modified in sequence by using the missing value ratio method,the discrete trend method and the R-type clustering analysis.Finally,thirteen segmentation variables which indicate the value of the industrial electricity customer are extracted.(3)A segmentation attribute dissimilarity metric model of the industrial electricity customer for hybrid attributes comprising hierarchical categorical attributes is established.For the widespread categorical attributes in the power industry which have hierarchical structure,the dissimilarity metric method is discussed.The target distance and its mathematic description is defined which covers the shortage of the existing metric on reflecting preferences information of the decision maker.A new dissimilarity metric is proposed which not only considers the conceptual similarities among each categorical attributes,but also reflects the differences in a particular decision target of them.An algorithm for the target distance's calculation is proposed by using the fuzzy similarity priority ratio decision and the tree breadth-first traversal method.A segmentation attribute dissimilarity metric model of the industrial electricity customer for hybrid attributes comprising hierarchical categorical attributes is established.The effectiveness of the model is validated by a simulation test.(4)An adaptive evolutionary clustering algorithm is presented for segmenting electricity customers dynamically.After analyzing the contractual characteristic of electricity customers and defining the concept of dynamic clustering,the basic principles and the algorithm of adaptive evolutionary clustering is mainly discussed.Determination criteria of cluster structure changes are designed to overcome the shortage of the Adaptive Forgetting Factor for Evolutionary Clustering and Tracking(AFFECT)on dealing with cluster structure changes and the problem of matching clusters between time steps.Under the determination criteria,the processing modules of eliminating clusters and adding new clusters are established and an adaptive evolutionary clustering framework is proposed.An adaptive evolutionary clustering algorithm under this framework for segmenting electricity customers dynamically based on the rough k-means clustering is also presented.(5)An application instance is presented to explain and validate the above segmentation model,segmentation variables,the dissimilarity metric model of hybrid attributes data and the dynamic clustering algorithm by using real data of industrial electricity customers stored in the database of a power supply bureau which subordinate Shaanxi Electric Power Company of China.