Physics-Consistency Enhanced Identification of Degradation Parameters in DC-DC Converter Devices

As power electronic devices become widely adopted in modern energy systems, their performance degradation increasingly affects system stability and service life. In DC-DC converters, aging of key components such as inductors, capacitors, and power switches leads to nonlinear changes in port responses, posing challenges for state estimation and fault prevention. To address this, we propose a modeling approach enhanced by physics-consistency mechanisms. By integrating differential constraints from converter circuits into neural network training, the model achieves improved physical reliability. Using measurable signals such as voltage and current, the method jointly minimizes data loss and physical residuals to estimate degradation-sensitive parameters like equivalent series resistance and capacitance. Experiments show that the proposed model maintains high prediction accuracy, robustness, and generalization under various degradation conditions, highlighting its potential in device health assessment and condition monitoring.