Recurrent Neural Network vs Kalman Filter for Development of Digital Twin of DC-DC Converters

This article presents a comprehensive comparison of model-based and data-driven state observers for dc-dc converters. Observers play a critical role in dynamic systems, providing insights into unmeasurable states for control and diagnosis. The objective of this study is to compare the development, accuracy, computational efficiency and applicability of model-based and data-driven observer approaches. As a case study, a Kalman filter will function as the model-based observer, and a recurrent neural network will function as a data driven observer for estimating the states of a bidirectional DC-DC converter in boost mode operation. The observers are implemented for simulated and physical boost converter models, and the results of the comparison are presented for each.