Abstract
This study introduces a novel Hybrid Random Forest-Symbolic Regression model for precise prediction of hourly solar irradiance in mountainous and urban regions. Leveraging geostationary satellite imagery and on-site measurements, the proposed model shows a good performance in accurately predicting solar irradiance, even under challenging conditions such as horizon effects and shading. Notably, it provides interpretable mathematical expressions, enhancing our comprehension of the underlying physical processes. Key findings reveal R2 values of 0.91 for Global Horizontal Irradiance (GHI) and R squaredvalues ranging from 0.88 to 0.91 regarding different cardinal orientations for Global Tilted Irradiance (GTI). Additionally, the model exhibits efficient usage of computational resources, featuring memory utilization between 340.35 MB and 1461.80 MB , execution times spanning 35.41 to 172.89 s, and CPU utilization rates from 26.6 % to 37.6 % during mathematical expression generation. This research has the potential to advance the development of environmentally friendly and efficient solar energy systems in terrain-affected regions, aiding in the design of energy-efficient buildings while promoting responsible resource usage.