Real-Time HIL Emulation of DRM With Machine Learning Accelerated WBG Device Models

Abstract

The proliferation of artificial intelligence (AI) has opened up new avenues for the modeling of power electronics with ultra-fast transient responses, such as wide-bandgap (WBG) devices. This article highlights the significance of ultra-fast transient device-level hardware emulation for the DC railway microgrid (DRM) in real-time. To this end, the proposed approach partitions the DRM power system by transmission line method (TLM) and employs gated recurrent unit (GRU) and electromagnetic transient (EMT) modeling techniques for system-level subsystems. Meanwhile, for WBG devices, gallium nitride (GaN) high electron mobility transistors (HEMT) and silicon carbide (SiC) insulated gate bipolar transistors (IGBT) are modeled using a novel physical feature neuron network (PFNN), which offers high flexibility with a variable time-step (as low as 1 ns), thereby improving the accuracy, efficiency and accelerating the emulation on the field-programmable gate array (FPGA). The effectiveness of the proposed approach is confirmed by comparing the emulation results with offline simulation results obtained from PSCAD/EMTDC for system-level and SaberRD for device-level transients. The proposed PFNN approach provides strong versatility, ultra-fast transient emulation capability, and significantly improved accuracy, which bodes well for the future of power electronics device-level emulation.

Keywords

• Artificial intelligence (AI)
• DC railway microgrid (DRM)
• field-programmable gate arrays (FPGAs)
• gallium nitride (GaN)
• gated recurrent units (GRU)
• hardware-in-the-loop (HIL)