Risk early warning of distribution power system based on data mining technology

In order to improve the accuracy of distribution network fault analysis and risk early warning research, in this paper, based on the data mining technology, the related factor analysis and risk early warning of distribution network are put forward. Through data cleaning, data transformation, data integration and outliers elimination, this paper sums up 27 kinds of distribution network fault features of the 4 categories. The improved Relief-Wrapper algorithm is used to analyze the fault correlation factors, 6 redundant features are eliminated, and the optimal fault feature subset composed of 21 fault features is formed. Taking into account the frequency of failure and the proportion of the loss of load, the distribution network fault risk indices and risk classification method is proposed. Using Random Forests Algorithm (RFA) method and optimal fault feature subset to carry on the risk early warning. Finally, an example of 120 feeder distribution network is given to verify the validity of the proposed method.