Meta Ensemble Learning with Acoustic Spectrum Analysis for Intelligent Diagnosis of Direct-buried Transformer Substations

This paper proposes a meta ensemble learning method to extract multiple electrical and non-electrical features, including acoustic signal, voltage, current and temperature, etc., for the status monitoring and fault diagnosis of direct-buried transformer substation (DBTS). The proposed method consists of a meta learning model and a sub-classifier model. In the meta learning model, a multi-input and multi-output back propagation neural network (BPNN) with multiple hidden layers is utilized to dynamically update the integrated weight coefficients of sub-classifiers based on the extracted multiple features of DBTS for accuracy improvements. The sub-classifier model involves four pre-trained classifiers using XGBoost for the fault type classification. Consequently, the fault type can be identified by weighted summation of integrated weights and classification results from the sub-classifier model. Comparative studies have been investigated to demonstrate the superior performance of the proposed meta ensemble learning method in improving the accuracy of DBTS fault diagnosis.