Abstract
This study addresses the difficulty of predicting triboelectric nanogenerator outputs across heterogeneous devices and reporting conventions. A dual-track framework is introduced: a closed-form surrogate grounded in Thevenin load matching that reconstructs the power-load relation and estimates internal resistance, baseline capacitance, and a charge-density proxy; and a machine-learning predictor trained with leave-one-study-out validation and equipped with uncertainty quantification. Using a curated, load-aware corpus of 20 independent studies (443 device conditions) harmonized to SI units, the framework predicts open-circuit voltage (V(oc)) and short-circuit current (I(sc)) with calibrated prediction intervals and study-level robustness, and it exposes actionable design rules such as optimal load near the internal resistance. The results connect device descriptors (area, excitation frequency, and surface treatment) to performance while supporting reproducible comparison across studies. This approach provides a practical pathway to physics-guided prediction and design of triboelectric systems for self-powered sensing and low-power energy harvesting.