Optimizing monthly solar PV tilt angles and energy yield across global climate zones: a hybrid machine learning and PVLib approach

Abstract

Accurate determination of photovoltaic (PV) tilt angles is essential for maximizing energy yield, yet traditional methods often rely on fixed or latitude-based empirical rules that overlook dynamic climatic factors. While machine learning (ML) has been applied for solar irradiance prediction and system optimization, few studies have directly targeted monthly tilt angle prediction. This paper presents a hybrid framework that integrates PVLib-based simulations with supervised ML models to predict monthly optimal tilt angles for fixed PV installations across diverse climate zones. The framework uses simulated energy output as ground truth and trains models using localized features such as solar radiation, ambient temperature, humidity, and geographic data. Applied across 17 cities spanning tropical, dry, temperate, and continental climates, the Random Forest surrogate reproduced PVLib-optimal monthly tilts with MAE 2.04 ± 0.04° and R2 0.975 ± 0.001 under 5-fold cross-validation (67,287 out-of-fold samples). Relative to Klein's fixed-tilt rule, ML-guided monthly tilts increased annual yield by a median 7.8 % (interquartile range 7.0–8.5 %), while the maximum gain was 12.2 % in Reykjavík. Statistical validation using paired t-tests and Wilcoxon signed-rank tests confirms the robustness of the approach. Results highlight ML's scalability and adaptability, offering a measurement-free alternative for climate-responsive PV design.