Machine Learning and Deep Learning Approaches for Optimal Solar Panel Tilt Angle Prediction Using Meteorological Data from Kano, Nigeria

The energy yield of photovoltaic installations depends critically on panel orientation, yet most installations in developing regions rely on fixed tilt angles chosen by rule of thumb. This study develops and compares classical machine learning and deep learning models for predicting the optimal hourly tilt angle of solar panels using multi-year meteorological measurements from the Kano region of Nigeria. A timestamped dataset of global horizontal irradiance, direct normal irradiance, diffuse horizontal irradiance, module temperature, wind speed, and relative humidity was cleaned, deduplicated, and enriched through feature engineering, including derivation of the clearness index from extraterrestrial irradiance, a fixed surface albedo of 0.23, solar zenith angle computation from site geometry, and specific humidity estimation from the Tetens relation. The target tilt angle was defined from the solar zenith angle. Four regression models were trained with cross-validation: linear regression, decision tree, random forest, and XGBoost. Hyperparameter tuning with randomized search showed XGBoost (RMSE 1.94) and random forest (RMSE 3.06) far outperforming linear regression (RMSE 564.25). A stacked long short-term memory network with 128 and 64 units, trained on 120-hour input windows with early stopping, achieved the best performance with an RMSE of 0.24, and test-set residuals largely fell within plus or minus 5 degrees. Predicted tilt profiles reproduced physically expected behaviour: diurnal peaks near solar noon, a seasonal cycle peaking at about 3.1 degrees in June and July and falling to about 1.7 degrees in December and January, and a consistent intra-month decline of roughly 4 degrees in January following the December solstice. The results demonstrate that sequence-based deep learning can capture both the orbital and atmospheric dynamics governing panel orientation and provide a foundation for intelligent, low-cost solar tracking in high-irradiance developing regions.