Adaptive Real-Time Hybrid Neural Network-Based Device-Level Modeling for DC Traction HIL Application

DC traction drive systems require high-frequency switching in the power converter whose device-level switching transients have a significant impact on the accuracy of hardware-in-the-loop emulation. Real-time device-level emulation has high computation demand for calculating the switch on and off transients. This paper introduces a new method to estimate the switching transients by utilizing artificial intelligence in the hardware design. In the hybrid neural network, the k-nearest neighbors (kNN) concept and the recurrent neural network (RNN) have been employed to emulate the transient waveforms in the DC traction drive. The kNN module classifies the switching states while the RNN module predicts the transient current for a specific condition. This work also proves that the classification of the input switching states with the help of kNN can play an essential role. The hardware implementation of the study case can be executed at a time-step of 100 ns with device-level transients. The results have been validated by PSCAD/EMTDC® at system-level and SaberRD® at device-level.