Power Customer Analysis And Power Outage Sensitivity Analysis Based On Machine Learning

With the advancement of the power system reform,the power grid is being transformed from a production-oriented enterprise to a marketing-oriented enterprise.At the same time,the release of the power-supply side has also brought certain operational pressure on the power supply company.The power supply company has changed from a pure sales company before.For companies that have a service nature.How to grasp the characteristics of users in the power market,occupy a large market share,how to determine the impact of power outages to users,and accurately depict users' sensitivity to power outages have become important issues for major power sales companies.This paper uses statistical analysis and machine learning related theories for user analysis and outage sensitivity analysis.First of all,statistical analysis of the data found that residents and non-residents have a lightweight 28 law between the number of users and the payment of electricity fees,and there are two-eight laws for large industrial electricity use and general industrial and commercial electricity consumption in non-resident electricity use;At the same time,the tariffs paid by electricity customers are consistent with the long tail theory.These three laws have very important guiding significance for how the power sales company formulates power sales and service strategies.The power sales company can maximize its capital by seizing important customers.Secondly,the datasets that are sensitive to power outages are correlated,analyzed,and clustered to identify variables that have a significant impact on power outage sensitivity.Undersampled datasets are used to solve under-fitting and overfitting problems;further improvement is made on the full data set.The detection performance of the decision tree is discussed,along with changes in the detection accuracy,detection coverage,false detection rate,missed detection rate,and harmonic number of the decision tree with the number of layers and fork trees,to find the optimal parameters,and discussed under this parameter.The ROC curve,trapezoidal table and line chart,etc.;finally compare the performance of the decision tree,SVM and Logistic regression model under optimal parameter selection,and found that the decision tree model is more suitable for predicting outage-sensitive users.