Design And Implementation Of Power Grid User Behavior Analysis System Based On Data Mining

With the development of the market economy, the civil electrical power industrygradually moves from production-oriented enterprises to business enterprises. Hencethe concept of marketing management enters the electrical power industry. Recently,with the rising of informatization of the power grid, lots of data accumulates in allkinds of electrical management systems.However, the existing systems can only simply operate the data. The deeprelations and rules implied by the data remain undiscovered, which can be used foranalyzing and predicting their tendencies. In view of the above situation, it is ourmain purpose to build a data mining system in order to analyze and predict userbehaviors.The project is based on analytical model research project of China SouthernPower Grid. The project aims to perform an advanced analysis to core data, whichinvolves electric quantity prediction, customer behavior mining, equipment failureanalysis, etc. After a series of theoretical and algorithmic researches of the team, thispaper concentrates on the implementation of the whole system, which is based on theStruts-Spring-Hibernate (SSH) framework and data mining technologies. The systemadopts the browser/server (B/S) architecture which targets at data mining on theelectricity customer segmentation, customer credit rating and prediction of arrearshigh-risk customers. According to the classic flow of data mining, the system consistsof five modules: data collecting, data preprocessing, data mining, results analysis, andsystem management. In the data preprocessing module, according to the continuous values and discrete values, I have implemented many methods including ignoring themissing data, deleting negative, constant value filled, the overall mean filling, fillingthe most likely value, and the regression filling for the data cleaning. I haveimplemented the methods including direct deletion, chi-square test, information gain,and information gain rate for the data reduction. And I have implemented the methodsincluding decimal scaling, maximum and minimum values, and Z-SCORE for thedata conversion. And the data mining modules, I have implemented many algorithmsincluding naive bayes classifier, ID3decision tree, analytic hierarchy process,K-Nearest Neighbor and K-means, which contribute to making detailed analysis andperditions.