Research On Residents’ Electricity Purchasing Behavior Based On Data Mining

The market-oriented reform of the power industry has carried out more than a decade,and power supply companies faced the combined pressure of external markets and internal management.In 2017,Beijing Electric Power Company has implement "full coverage and full collection" of smart meters throughout the city.The popularization of smart meters has made it possible to acquire real-time electricity purchase information from residents.Massive data makes the management of electricity marketing services enter the era of big data.The key to improving the competitiveness of power supply companies is using massive data and data mining techniques to analyze the electricity purchasing behavior of residents,so it is necessary to conduct research on the characteristics and changes of residents’ purchase behavior and provide reference for the business management of power supply companies.First of all,Based on customer behavior analysis theory and data mining in the power industry and other fields.I also analyze the electricity purchasing process of residents,extract fields reflecting residents’ attributes and property purchasing behavior from the four databases(residents’ unified platform,electricity sales management system,marketing management system,and collection and delivery system),and sort out and reduce these attributions.Second,taking into account the distribution of residential areas and the use of electricity,a stratified sampling system was used to extract 8,000 households from more than 1.3 million smart meter households in the city.The residents’ purchase records from 2013 to 2017 were used as the sample.After the sampling is completed,the sampling data is pre-processed,including removing duplicate data values,removing invalid data values,and replenishing blank data.Then,the multiple variables used in the analysis are classified according to the meaning they represent,and they are divided into four dimensions:time dimension,behavior dimension,geographic dimension and other dimensions.Through multi-dimensional analysis,we have obtained changes in the perspectives of residents,purchase time,payment,purchase frequencies,channels,and methods for purchasing electricity.We also have built comprehensive understanding of the laws of electricity purchase,the differences in the purchase of electricity between regions or levels,and opinions for the intermal optimization of the company.Finally,the K-means clustering algorithm was used to select two clustering features:payment and frequencies of electricity purchased.I clustered the residents of 2016 and 2017,and divided the residents into 5 categories,providing the company a better way to conduct business analysis and channel management based on these four types of residents’ different purchasing habits.I define and program the residents’ purchasing power channel preference and power purchasing preference,and predict the proportion of three channel preference residents from 2018 to 2020 using the 2016 and 2017 data to construct the Markov transition matrix,using the Apriori correlation algorithm to obtain the relationship between residents’ attributes and power purchase behaviors by selecting the clustering category of residents,electricity subscale and area category as pre-excavation,power purchase methods and power purchase channels as the post-excavation items.