Photovoltaic tilt optimization using physics-informed machine learning methods

Photovoltaic arrays are typically installed at a tilt angle, measured from the horizontal plane, close to the site latitude; however, this approach can underperform relative to the physics- or data-driven optimum tilt, resulting in suboptimal energy yields. In this work, we introduce a physics-informed machine learning framework to determine the annual fixed tilt that maximizes energy yield for Antalya, Turkiye. To this end, we process daily irradiance records to derive the diffuse fraction and seasonal harmonics. We use the Liu-Jordan isotropic model to generate a physics-based reference curve and obtain an annual optimum array tilt of 34.75 ∘ . Afterward, we apply machine learning algorithms to a tilt-augmented dataset (0-60 ∘ , in steps of 0.25 ∘ ). The best-performing machine learning model yields an incident irradiation of 1 , 850 kWh m − 2 year − 1 , which corresponds to an electrical specific yield of 1.57 MWh kWp − 1 year − 1 (assuming a performance ratio of 0.85). Finally, all the results provide a robust way to select the tilt angle for optimal system design and to compare data-driven with physics-based models.