Author(s) :

Nihal Kadamba, Mauricio Hernandez.

Volume/Issue :

Abstract :

The increasing use of renewable energy sources such as solar and wind power has led to significant challenges with energy curtailment, causing substantial losses in energy efficiency. To address this, we investigated the use of machine learning models to predict solar energy output and mitigate curtailment across California, Nevada, Arizona, and New Mexico from 2018 to 2022. We hypothesized that time of the day, temperature, and irradiance would be the factors most predictive of energy curtailment across the aforementioned states. Specifically, we used XGBoost and Random Forest to predict solar curtailment and evaluate if such predictors can be used to accurately estimate curtailment in each state. Using historical weather, the characteristics of solar power plants, and energy data, we found that these models effectively predicted general energy patterns. Notably, we found that the hour of the day and the average temperature across all solar plants in each state were the main predictors in the models. Using a set of 16 predictors, the evaluated models reported mean absolute errors ranging from 1.8% to 4.8% of the historical solar energy curtailed. These results highlight the potential of machine learning to optimize renewable energy use, reduce curtailment, and improve grid reliability, offering a scalable solution to address the challenges of oversupply in renewable energy systems.