Thermal Storage System Optimization: 11% Energy Cost Reduction

Thermal storage systems that control cooling in large buildings exhibit varying energy usage based on their operational methods. Sudden control actions, often triggered by temperature complaints, lead to energy usage spikes and increased operating costs. Current control scenarios, based on the experience of on-site operators, result in significant deviations due to varying levels of proficiency. There is a need to optimize control methods that accurately reflect the actual state and data of the thermal storage system.

An AI-based simulator was created using a regression model of the facility and historical data. Control variables were screened through data cleansing and exploratory data analysis (EDA) to ensure the reliability of the data from the thermal storage system. Reinforcement learning agents were utilized to generate optimized control strategies for thermal storage systems, providing on-site managers with data-driven control scenarios.

The reinforcement learning-based control method reduced the average daily electrical energy cost by about 11% compared to conventional methods. By providing data-driven optimal control strategies, we minimized deviations among site managers, moving away from reliance on personal experience to more consistent and efficient operations.