Abstract:Distribution feeder fault prediction is an important way to improve the distribution network reliability, in order to construct a feeder fault prediction model with stable performance and strong forecasting ability, it is necessary to ensure the validity, strong correlation and non-redundancy of the model input feature variables.For reasonable determine the characteristics of input variables of the feeder fault prediction model, this article adopts the analysis method of data exploration and mining of the relationship between the feeder fault and its influencing factors are analyzed, Specifically, through correlation analysis of a large number of actual feeder fault data based on Pearson Correlation Coefficient, The correlation statistics of large number of actual feeder fault data and its influencing factors were carried out. From four dimensions of feeder fault such as time-regional characteristics, external influencing factors, self-influencing factors, and operation influencing factors, the characteristic variables of feeder fault influence factors are selected as input variables of feeder fault prediction model,The independent fault feature variables are eliminated intuitively and effectively.Therefore, the proposed method can be used for pretreatment analysis and extraction of big data of distribution network, which provides important method and data base for distribution network fault prediction.