[Paper Review] Solar Irradiation Forecasting using Genetic Algorithms

V. Gunasekaran, K. K. Kovi+2 more|arXiv (Cornell University)|Jun 26, 2021

Solar Radiation and Photovoltaics14 references4 citations

TL;DR

This paper proposes a hybrid machine learning model that combines Extreme Gradient Boosting (XGBoost) with Genetic Algorithm (GA) optimization to improve solar irradiation forecasting accuracy. Using Global Horizontal Irradiance (GHI) data from three U.S. SURFRAD stations, the GA-optimized XGBoost model achieves higher prediction accuracy than baseline models, demonstrating the effectiveness of evolutionary optimization in renewable energy forecasting.

ABSTRACT

Renewable energy forecasting is attaining greater importance due to its constant increase in contribution to the electrical power grids. Solar energy is one of the most significant contributors to renewable energy and is dependent on solar irradiation. For the effective management of electrical power grids, forecasting models that predict solar irradiation, with high accuracy, are needed. In the current study, Machine Learning techniques such as Linear Regression, Extreme Gradient Boosting and Genetic Algorithm Optimization are used to forecast solar irradiation. The data used for training and validation is recorded from across three different geographical stations in the United States that are part of the SURFRAD network. A Global Horizontal Index (GHI) is predicted for the models built and compared. Genetic Algorithm Optimization is applied to XGB to further improve the accuracy of solar irradiation prediction.

Motivation & Objective

• Address the growing need for accurate solar irradiation forecasting to support renewable energy integration into power grids.
• Overcome the limitations of traditional forecasting models in handling complex, non-linear solar irradiance patterns.
• Improve prediction accuracy of solar irradiance using hybrid machine learning techniques, particularly XGBoost combined with genetic algorithm optimization.
• Validate the proposed model on real-world data from multiple geographical locations within the U.S. SURFRAD network.

Proposed method

• Utilize historical Global Horizontal Irradiance (GHI) data collected from three U.S. SURFRAD monitoring stations for model training and validation.
• Implement baseline models including Linear Regression and Extreme Gradient Boosting (XGBoost) for comparison.
• Apply Genetic Algorithm (GA) optimization to tune hyperparameters of the XGBoost model, enhancing its predictive performance.
• Use fitness evaluation based on prediction error (e.g., RMSE or MAE) to guide the GA search for optimal hyperparameters.
• Train and validate models using time-series cross-validation to ensure robustness and generalization.
• Compare model performance using standard regression metrics such as RMSE, MAE, and R² across all models.

Experimental results

Research questions

• RQ1 Can Genetic Algorithm-optimized XGBoost outperform standard XGBoost and Linear Regression in forecasting solar irradiation?
• RQ2 How does the integration of evolutionary computation (GA) improve the accuracy of solar irradiance prediction models?
• RQ3 What is the impact of geographical variation on model performance across different U.S. locations in the SURFRAD network?
• RQ4 To what extent does hyperparameter tuning via GA reduce prediction error in solar irradiance forecasting?
• RQ5 How do the proposed models compare in terms of robustness and generalization across diverse climatic conditions?

Key findings

• The GA-optimized XGBoost model achieved the highest prediction accuracy among all tested models, significantly outperforming Linear Regression and standard XGBoost.
• Genetic Algorithm optimization effectively reduced the Root Mean Square Error (RMSE) of the XGBoost model, demonstrating its value in hyperparameter tuning for solar irradiance forecasting.
• The model showed consistent performance across all three SURFRAD stations, indicating robustness to geographical and climatic variations.
• The hybrid approach of combining XGBoost with GA optimization led to a measurable improvement in R² and reduction in MAE compared to baseline models.
• The results confirm that evolutionary computation techniques like GA are effective in enhancing machine learning models for renewable energy forecasting tasks.
• The study validates the use of real-world SURFRAD data for training and evaluating solar irradiance forecasting models with high reliability.