Advanced solar radiation decomposition model: utilizing Random Forest-Symbolic Regression across diverse climatic zones

Solar radiation decomposition, the process of separating global horizontal irradiance (GHI) into direct and diffuse components, is essential for applications such as building performance simulations and solar energy systems. Traditional models like Liu-Jordan, Erbs, Reindl, Boland-Ridley-Lauret, and Perez often struggle with accuracy across diverse climatic zones due to their dependency on local atmospheric conditions.

This study addresses these limitations by proposing a novel method utilizing a Random Forest-Symbolic Regressor (RF-SR) to develop a more adaptable and accurate hourly time-step all-sky solar decomposition model. RF-SR combines the ensemble learning capabilities of Random Forests, which handle large datasets and complex interactions between GHI and atmospheric parameters from satellite imagery and the interpretability of Symbolic Regression, to finally generate concise mathematical expressions to estimate direct and diffuse components of solar radiation.

The research utilizes a comprehensive dataset covering various geographic locations and climates across the world, including tropical, dry, temperate, and continental, to train and test the performance of the proposed RF-SR model against traditional ones. Performance is evaluated using statistical metrics, including the coefficient of determination (R²), mean absolute error (MAE), mean bias error (MBE) and root mean square error (RMSE).

We anticipate that the RF-SR model outperforms established models in accurately decomposing GHI across different climatic zones. The direct and diffuse irradiance data from the analyzed decomposition models are integrated into EnergyPlus simulation software through a streamlined workflow. This approach uses Python to dynamically input DNI and DHI data into the EnergyPlus weather file, allowing us to assess the impact of varying irradiance accuracy on domestic PV system performance.

It enables a detailed evaluation of how solar conditions, including the impact of shading, orientation, and local climate, affect the system’s contribution to building energy efficiency and renewable energy integration. The anticipated findings suggest that the RF-SR model offers a more accurate and broadly applicable solution, benefiting renewable energy management and sustainable building design.