Data-Efficient Physics-Informed Learning to Model Synchro-Waveform Dynamics of Grid-Integrated Inverter-Based Resources

TL;DR

This work addresses the challenge of accurately modeling fast sub-cycle inverter-based resource transients with limited disturbance data. It introduces a data-efficient physics-informed learning framework that maps the IBR terminal current $\Delta i_1(t)$ from differential disturbance $\Delta v_1(t)$ while enforcing the grid-side Kirchhoff relation $\Delta v_2(t)=\Delta v_1(t)-R\Delta i_1(t)-L\dfrac{d}{dt}\Delta i_1(t)$, using two neural networks and either known $(R,L)$ or jointly learned $(R,L)$. The approach demonstrates superior accuracy and markedly improved data efficiency compared to purely data-driven baselines across multiple sampling rates, with the unknown-parameter case also enabling identification of $R$ and $L$. The results suggest a practical path toward reliable, data-efficient transient modeling for grid-integrated IBRs using WMU data, with extensions to more complex networks and measurement conditions planned for future work.

Abstract

Inverter-based resources (IBRs) exhibit fast transient dynamics during network disturbances, which often cannot be properly captured by phasor and SCADA measurements. This shortcoming has recently been addressed with the advent of waveform measurement units (WMUs), which provide high-resolution, time-synchronized raw voltage and current waveform samples from multiple locations in the power system. However, transient model learning based on synchro-waveform measurements remains constrained by the scarcity of network disturbances and the complexity of the underlying nonlinear dynamics of IBRs. We propose to address these problems by developing a data-efficient physics-informed machine learning (PIML) framework for synchro-waveform analytics that estimates the IBR terminal current response from only a few network disturbance signatures. Here, the physics of the electrical circuits are used to compensate for limited data availability by constraining the learning process through known circuit relationships. Two cases are considered, with known and unknown circuit parameters. In the latter case, the framework jointly learns the transient dynamics of the IBRs and the parameters of the electrical circuit. Case studies using WMU disturbance data across multiple sampling rates shows consistently lower current estimation error with substantially fewer training events than a purely data-driven baseline.

Figures (4)

• Figure 1: Grid-connected IBR monitored by GPS time-synchronized WMUs, showing the transient voltages $v_1(t),v_2(t)$ and current response $i_1(t)$ following grid disturbances across line with resistance $R$ and inductance $L$.
• Figure 2: The MSE for the testing data using data-only model (solid blue) and the PHIL model (dashed red) for known circuit parameters; see Section \ref{['subsec:\n knownRL']}. The PIML model achieves lower prediction error than the data-only baseline. (a)-(c) correspond to sampling rates of 128, 64, and 32 samples per cycle.
• Figure 3: The MSE for the testing data using data-only model (solid blue) and the PIML model (dashed red) for unknown circuit parameters; see Section \ref{['subsec:\n unknownRL']}. The PIML model achieves lower prediction error than the data-only baseline. (a)-(c) correspond to sampling rates of 128, 64, and 32 samples per cycle.
• Figure 4: The learned resistance $R$ and inductance $L$ over the training iterations with 20 training events. Both parameters are updated through gradient-based optimization and gradually converge to true parameters.