Conference Paper
Proceedings of SimBuild Conference 2024
Development of a Reinforcement Learning-Based Solar Decomposition Model for Predictive Control Using Limited Measurement Data
Development of a Reinforcement Learning-Based Solar Decomposition Model for Predictive Control Using Limited Measurement Data
Byung-Ki Jeon, Deuk-Woo KimDepartment of Building Energy Research / KICT, Korea, Republic of (South Korea)Abstract: The perfect prediction of Diffuse Horizontal Irradiance (DHI) for the following day is essential for maintaining a stable power supply and minimizing energy losses. In this study, the proposed model predicts the DHI for an entire year using only two weeks of DHI data and readily obtainable meteorological parameters. The model proposed in this research maintains the reliability of the physical equations inherent in existing decomposition models while incorporating reinforcement learning to enhance error reduction. The model proposed has demonstrated the capability to perform long-term solar decomposition calculations without the need for further model updates. Keywords: Solar irradiance, Decomposition model, Reinforcement learningPages: 598 - 606 Paper: