Deep learning-based estimation technique for capacitance and ESR of input capacitors in single-phase DC/AC converters

• Published in: JOURNAL OF POWER ELECTRONICS Vol. 22; no. 3; pp. 513 - 521
• Main Authors: Park, Hye-Jin, Kim, Jae-Chang, Kwak, Sangshin
• Format: Journal Article
• Language: English
• Published: Singapore Springer Singapore 01.03.2022; 전력전자학회
• Subjects: Electrical Machines and Networks; Engineering; Original Article; Power Electronics; 전기공학; Deep learning; Capacitance estimation; ESR; DC/AC converter
• ISSN: 1598-2092; 2093-4718
• DOI: 10.1007/s43236-021-00366-x

This study proposes an algorithm to estimate the state of an input capacitor based on a deep neural network (DNN). This algorithm runs in a DC/AC single-phase converter. According to the analysis result of the data from the capacitor, the component with twice the fundamental and switching frequencies demonstrated dominant characteristics. The most dominant low-frequency and mid-frequency components are extracted from the collected experimental voltage and current through a fast Fourier transform. With these four components, 11 combinations of input variables were created and used as inputs for the DNN. After training and testing, we determined which combination had the best performance. Therefore, in the case of capacitance, the use of a mid-frequency component together shows better performance than a low-frequency component alone. In the case of an equivalent series resistor, using both the mid-frequency capacitor voltage and the capacitor current component shows better performance than otherwise.