A Black-Box Modeling Method Based on LSTM Network for DC-DC Converters

Modeling power electronic converters is of great importance for verifying the stability and power quality of the power systems containing multiple power electronics devices. However, due to complicated inner structures, protected intellectual property, or other reasons, fidelity of system-level modeling and simulation is difficult to be ensured. A data-driven black-box modeling method for DC-DC converters powering pulsated power loads is presented based on the Long Short-Term Memory (LSTM) network. The complicated operation conditions of the input voltage and load current of converters are analyzed and firstly targeted for modeling. Then, the structure and hyper-parameters of the presented model is designed. Finally, the dataset for training, validation and test is obtained by sampling the voltages and currents of the target converter and then used for modeling. The fidelity of the DC-DC converter's model under more general operation conditions can be improved by the presented black-box model. The feasibility and accuracy of the presented method is verified by dataset acquired in experiments.