Fault Section Location Method Based on Random Forest Algorithm for Distribution Networks with Distribution Generations

The fault information collected by the feeder terminal units is easily distorted under the undesirable environment. The traditional fault section location methods of the distribution network are facing problems of poor fault tolerance and low accuracy. In this paper, a novel fault section location method for distribution networks is proposed based on the random forest algorithm. Considering the high-dimension and unbalanced characteristics of the original fault sample sets, the SMOTE oversampling method and the learning-model-based feature selection method are utilized to preprocess the original data. Furthermore, for establishing the fault location prediction model, the relevance between the fault attributes and the fault section is mined by the random forest algorithm. The simulation results show that the proposed method can achieve the single and multiple fault section location in distribution networks. Meanwhile, compared to the traditional methods, the proposed method has two significant advantages, strong fault tolerance to distorted information and high prediction accuracy.