Optimizing Cooling in Modular Data Centers: A Machine Learning Approach Using Real-Time Thermal Imaging

The current volume of electricity used worldwide results in significant greenhouse gas emissions and the release of other air pollutants into the environment. Data centers are major power consumers due to the substantial energy required for computation and cooling, which is necessary to prevent servers and equipment from overheating. If the ambient temperature is not regulated and maintained within the manufacturer’s specified range, electronic devices may malfunction, become damaged, or even be destroyed [1]. Using thermal images to forecast temperatures in a modular data center can reduce the duration of active cooling and its associated environmental impact. Targeted cooling is expected to lower the data center’s Power Usage Effectiveness (PUE) while optimizing the energy needed to maintain operational functionality. A dataset with over 1,152,000 thermal images, captured at intervals of every 6 seconds, within the modular data center over six months. The data is initially processed using a convLSTM model to train and predict future images and temperature patterns. The goal is to achieve a prediction accuracy of at least 85%, ensuring the model's reliability for real-world implementation. The active cooling reduction will be measured to determine the percentage decrease in energy use compared to baseline operation. This research aims to promote sustainable practices by reducing active cooling and incorporating external environmental temperatures, combined with predictive thermal imaging models. This predictive approach also has the potential to be applied in other contexts, such as food and medication storage, where precise temperature-controlled cooling is essential.