Work

A Physics-Informed Self-Calibrating Digital Twin with Conformal Uncertainty for Multi-Inverter PV Power Systems

Year: 2026
Type: conference-paper
Abstract: This research introduces a physics informed, self-calibrating digital twin for multi-inverter photovoltaic (PV) power systems and uncertainty quantification frameworks of any uncertainty type that satisfy the conformal type assumption. A "compact" irradiance-temperature physics baseline model, an inverter-aware residual correction model, conformal physics "normalized" prediction intervals, and an online self-calibration mechanism for the inverter-specific operating state in the sequence deployment are integrated within the proposed framework. The proposed approach is not only created to significantly enhance the accuracy of PV forecasting, but also to reveal abnormal behavior of the inverter at the inverter level using the uncertainty-aware screening technique. Evaluation of the method is done for the inverters level PV plant data, including 22 inverters, operating in a realistic PV plant environment. The results from the experimental test indicate that the offline hybrid model has a lower mean absolute error of 398.5 W than the other models, while the online self-calibrating digital twin has the best overall performance among the three models with R2=0.9155, RMSE =794.1 W and PICP =95.34% and the lowest WIS of 4304.9 W. Additionally, the uncertainty analysis also highlights that a subset of inverters (confined to a small window) continues to demonstrate under coverage and high risk, even with good fleet-level results. The results illustrate how self-calibration with uncertainty can enable a hybrid PV predictor to become a viable digital-twin system for fleet monitoring, prioritization of inverters and operational decision-making.
Language: English
Field: Energy
Open Access status: hybrid
OpenAlex ID: W7214639732
FWCI:
Cites: 13
Cited by:
Related to: 10