A Data-driven Approach with Improved Generalization Performance to Modeling Transient Behaviors of DC-DC Converters

Nowadays, the data-driven approaches to modeling power electronics (PE) systems are mostly based on sequential neural networks (NNs). These approaches may require too much data since the NNs can not generalize across a wide range of inputs. To address this issue, this paper proposes a new data-driven approach to modeling the transient behaviors of DC-DC converters, which is based on fully-connected NNs. The proposed method introduced prior knowledge about linear systems and thus significantly improved the generalization performance. In this method, circuit parameters are first mapped into linear system characteristics by fully-connected NNs, and then the outputs are calculated by the inputs and the system characteristics. Experiment results show that the entire circuit topology with configurable parameter settings and initial conditions can be successfully modeled. Parameter change events are also supported by this approach.