Accuracy enhance of fault classification and location in a smart distribution network based on stacked ensemble learning

Abstract

This paper proposes an ensemble learning-based method to determine the type and location of faults in a smart distribution network. The employed scheme is a stacking based method that consists of two main levels, each of which includes one or more independent classifiers. In the structure of the proposed method for first level, multiclass support vector machine, K-nearest neighbors and random forest models are exploited. Also, for the next (final) level, which is also called meta-classifier, a random forest based model has been used, which itself is a kind of ensemble learning method. In order to supply the data used to feed the model, the voltage and current measurements in some buses and lines have been used. The proposed model is tested using the data related to all different types and location of faults and operating conditions in the IEEE 123-bus network. Besides, the model is implemented on another distribution test system to consider wider range and specs for fault parameters and also to comprise with other methods applied to it; high impedance faults are also regarded in this part. The obtained results show the appropriate accuracy of the proposed model and its better performance than other considered methods.