A Highly Accurate Generative Learning‐Based DC‐Link Capacitance Estimation Approach for Electrified Railway Traction Systems

ABSTRACT

To enhance the reliability of electrified railway traction systems (ERTSs), timely replacement of the DC‐link capacitor is essential. Unfortunately, current DC‐link capacitance estimation methods are often characterized by low accuracy and poor robustness to measurement noise. To address these issues, this paper presents an innovative DC‐link capacitance estimation method that achieves high accuracy with a small experimental dataset. The variational autoencoder (VAE) algorithm, a generative machine learning technique, is used to effectively expand the dataset by generating new data, significantly reducing the cost and time associated with extensive experimentation. Additionally, the long short‐term memory (LSTM) algorithm is employed to estimate the DC‐link capacitance in ERTSs with high precision and strong resistance to measurement noise. The effectiveness of the proposed approach is validated through experimental results and analysis.To enhance the reliability of ERTSs, timely replacement of the DC‐link capacitor is essential. This paper presents an innovative DC‐link capacitance estimation method that achieves high accuracy with a small experimental dataset, which addresses the problems of the existing capacitance estimation approaches, i.e., low accuracy, poor robustness to noise and/or the requirement of a large amount of data.