Intelligent Data-Driven Digital Twin for Predictive Health Monitoring of 3-Phase Grid-Connected PV Systems

Abstract

Grid-connected photovoltaic (PV) inverters are increasingly exposed to dynamic irradiance variations, distorted grid voltages, and progressive passive-component degradation, which can reduce power quality, increase losses, and compromise long-term reliability. Conventional physics-based digital twins require detailed analytical models and continuous synchronization with the physical system, while many diagnostic methods remain offline or depend strongly on predefined component models. To address these limitations, this paper proposes a data-driven Digital Twin (DD-DT) framework for state estimation, internal parameter identification, and predictive health monitoring of a single-stage three-phase grid-connected PV inverter. A Cascaded Second- and Third-Order Generalized Integrator-Noise Suppression Block-Compact Generalized Integrator (CSTOGI-NSB-CGI) is used for robust voltage-template extraction under distorted grid conditions. An Echo State Network (ESN) is then employed as a reservoir-computing surrogate to learn nonlinear temporal relationships between measurable signals and hidden internal electrical states. The reconstructed states are further used to estimate critical component parameters, including interfacing inductance, capacitance, equivalent series resistance, and switch on-state resistance. Experimental validation under progressive interfacing-inductor and DC-link-capacitor degradation confirms close agreement between hardware and DD-DT responses, with relative prediction errors maintained within 3%, an overall average $\text{NRMSE}$ of 5.57%, and an average $R^{2}$ of 0.876 across the reported test cases. The developed framework provides a scalable and computationally efficient basis for predictive maintenance of grid-connected PV inverter systems within Industrial Cyber-Physical Systems (ICPS).