Inverter Dynamic Response Simulation Via Neural Ordinary Differential Equations

The paper focuses on simulating inverter dynamic responses using neural ordinary differential equations (NODE). It addresses the need for dynamic, data-driven models of Distributed Energy Resources (DER) inverters in power systems. Traditional modeling methods, such as switching and averaged linear models, have limitations either in computational complexity or capturing non-linear dynamics. The NODE method offers a novel solution by integrating neural networks with differential equations, allowing for continuous-time processing and adaptability. The paper demonstrates that NODE provides a more efficient, yet accurate method for modeling inverter dynamics, especially in handling nonlinearities, compared to the multiple transfer function method, with about 30% reduction in the modeling errors.