Sparse Identification of Nonlinear Dynamics for Digital Twin Applications in Solar Photovoltaic Systems

Abstract

Digital Twin (DT) applications for solar photovoltaic (PV) systems require dynamic models that are compact, interpretable, and capable of adapting to changing environmental conditions. This paper investigates the use of Sparse Identification of Nonlinear Dynamics with control (SINDYc) as a data-driven modeling approach to support DT applications in solar PV systems. Considering a PV panel directly connected to a resistive load, the voltage and current are algebraically constrained, motivating a reduced-order formulation in which the output current is treated as the sole dynamic state, while solar irradiance and module temperature are modeled as exogenous inputs. Synthetic datasets with diverse excitation profiles are first employed to identify sparse nonlinear dynamic models. A key contribution of this work is the application and validation of the SINDYc framework using experimentally acquired real-time PV data, demonstrating its effectiveness beyond synthetic or simulated settings. Model performance is evaluated using error metrics to assess predictive accuracy and generalization across operating conditions. The results demonstrate that SINDYc can identify reduced-order dynamic models capable of capturing the effective behavior of the PV system, thereby supporting core DT applications such as real-time prediction, monitoring, and model updating. Limitations related to generalization under unseen real-world conditions are discussed, outlining directions for future improvement of DT fidelity.