This project explored whether reinforcement learning could reduce the energy used to cool telecom site containers without allowing temperatures to drift beyond safe operating limits. Using a fitted thermal model together with a Deep Q-Network controller, the study showed that a learning-based cooling strategy could outperform conventional automatic control in both humid and arid climate scenarios.

The key finding was that cooling, which accounted for a large share of site energy use, could be managed more intelligently while still protecting sensitive equipment. In simulation, the learned controller used substantially less energy and triggered far fewer temperature alarms than the baseline rule-based strategy, pointing to a practical path toward lower-energy thermal management for mobile network infrastructure.