Modeling Power Electronic Converters Using A Method Based on Long-Short Term Memory (LSTM) Networks
Pouria Qashqai, Kamal Al‐Haddad, Rawad Zgheib
Abstract
Power electronic converters are the crucial components of numerous applications. Due to the switching nature of power electronic converters, they behave nonlinearly. Thus the detailed modeling of such converters imposes a huge complexity and calculation burden on their simulation. This complexity grows exponentially when the detailed models are integrated into the simulation of large networks. In this paper, a large signal modeling technique based on long-short term memory networks is proposed. The black-box approach enables the method to model commercial over the shelf converters. The size of the network provides enough degree of freedom for tuning a trade-off between complexity and accuracy. A DC/DC buck converter is modeled using this technique to validate its performance. The training datasets are obtained from the MATLAB/Simulink switching model of the converter. A long-short term memory network (LSTM) is then trained using MATLAB's deep learning toolbox. The obtained results demonstrate the performance of the proposed method over conventional black box modeling techniques based on neural networks.