Log-transformed Red-Tailed Hawk optimization for single-diode photovoltaic parameter extraction

Abstract

Accurate extraction of the five parameters of the single-diode photovoltaic equivalent-circuit model is hampered by the nine-orders-of-magnitude dynamic range of the diode saturation current I 0 , which can cause gradient-free optimisers to converge to degenerate solutions when I 0 is searched on a linear scale. This paper proposes a log 10 ( I 0 ) reparametrisation that maps the admissible saturation-current range uniformly onto a bounded interval, paired with a Newton–Raphson implicit current solver and a fixed-seed reproducibility protocol, with the Red-Tailed Hawk Algorithm (RTH) as the optimisation engine. The method is evaluated on three standard benchmarks: the Shell SM55 monocrystalline module, the RTC France 57 mm silicon solar cell, and the Photowatt-PWP201 module. RTH is compared against PSO, GWO, WOA, HHO, AVOA, and MGO over 30 independent runs. To avoid an unfair comparison caused by the three update phases of RTH, a budget-neutral experiment is also performed with exactly 149,050 objective-function evaluations per algorithm and run. Under this equal-FE protocol, RTH obtains the lowest median RMSE on SM55 (2.4895×10 −5 A), RTC France (7.7301×10 −4 A), and PWP201 (2.0530×10 −3 A) among the algorithms evaluated under the adopted protocol. An ablation study confirms that log 10 ( I 0 ) sampling substantially improves run-to-run robustness on RTC France, where linear I 0 sampling produces high-error outlier runs. The results support the proposed workflow as a reproducible implementation framework rather than as a fundamentally new PV model or a certified global optimiser.