Hybrid Feature Selection Approach for Power Transformer Fault Diagnosis Based on Whale Optimization Algorithm and Extreme Learning Machine

Abstract

To further improve fault diagnosis performance, a new hybrid feature selection approach combined with whale optimization algorithm and extreme learning machine is presented in this study. Firstly, three filter methods based on different evaluation metrics are employed to select and rank 25 input features derived from gases concentration values, gases ratio, and energy-weighted dissolved gas analysis. Then, feature fusion approaches are applied to aggregate feature ranks and form a lower-dimension candidate feature subset. Afterwards, the whale optimization-based extreme learning machine model is implemented to optimize parameters and select optimal feature subsets. The accuracy of the model is used to evaluate the fault diagnosis capability of the concerned feature subsets. Finally, novel subsets are determined as the optimal feature subset to establish a fault diagnosis model. According to the experimental results, the average accuracy of the proposed approach is better than that of other conventional methods, which indicates that the optimal feature subset obtained by the proposed method can significantly promote the fault diagnosis accuracy of the power transformer.