PV optimization using solar radiation prediction based on multi-region dataset

Accurate GHI data are essential for designing cost-effective PV systems, yet many regions lack on-site measurements. While recent studies have demonstrated that multi-region GHI prediction models generalize well beyond their training locations, these advances have not been leveraged for PV system optimization, which still relies on measured data, with traditional PV optimization approaches continuing to use observed GHI data.

In this study, we bridge that gap by employing a multi-region dataset to drive an ML-based GHI prediction framework, integrating its outputs into a five-stage PV performance model and then applying NSGA-II for multi-objective optimization of annual energy yield and LCOE.

The optimal PV configuration had a 39° tilt, 212° azimuth and module type 3 and achieved an LCOE of 0.1545 KRW/kWh. Differences from the measured based optimization were under 1° in tilt, 3° in azimuth and 6.6% in LCOE.

This framework enables cost-effective PV system design, even in regions with limited solar and PV data.