Sub-Kelvin Prediction of Photovoltaic Module Temperature with a Physics-Informed Stacked Ensemble Digital Twin, and Its Use to Quantify Nano-Enhanced Passive and Active Cooling

Authors

  • Merwa Alhadrawi Author

DOI:

https://doi.org/10.64751/tq0z6353

Abstract

Silicon solar-cell efficiency falls by roughly 0.4 % per kelvin of cell temperature rise, and field modules routinely run 25–35 K above ambient. Passive, active and nanotechnology-based cooling recover part of this loss, but reported benefits come almost entirely from short prototype campaigns that cannot be extrapolated to plant scale. This work builds a deployable thermal digital twin and uses it to run counterfactual cooling experiments on real plant data. In the proposed Physics-Informed Stacked Ensemble Network (PI-SEN), a non-negative blend of steady-state thermal models supplies a physical prior and a stacked ensemble predicts only the residual; thermal inertia enters through a bank of discrete resistance–capacitance filters applied to irradiance and ambient temperature, so no past module temperature is ever an input and the model is free of target leakage. On the public two-plant Solar Power Generation benchmark PI-SEN attains the lowest error in three of four plant-protocol combinations against thirty-six baselines, reaching 0.964 °C under a strict chronological hold-out and 0.878 °C under the random hold-out used by prior work — below the only unambiguous kelvin-scale benchmark in the literature. On the one exception, a monsoon-onset block outside the training distribution, a linear model proves more robust, which we report openly. Characterising each cooling technology by its heat-transfer enhancement shows that a hybrid nano-PCM plus nanofluid loop lowers the mean daytime module temperature by 11.0 K and recovers 4.94 % of net DC energy after the pumping penalty, against 3.72 % for passive cooling.

Downloads

Published

2026-09-28

How to Cite

Merwa Alhadrawi. (2026). Sub-Kelvin Prediction of Photovoltaic Module Temperature with a Physics-Informed Stacked Ensemble Digital Twin, and Its Use to Quantify Nano-Enhanced Passive and Active Cooling. International Journal of AI Electrical Civil and Mechanical Engineering, 2(3), 705-716. https://doi.org/10.64751/tq0z6353