一、中期答辩题目：DATA CENTER ENERGY COST MINIMIZATION TECHNIQUES THROUGH EFFECTIVE UTILIZATION OF RENEWABLE ENERGY SOURCES USING DEEP LEARNING AND REINFORCEMENT LEARNING
二、中期答辩人：Abubakar Rafique
三、中期答辩时间：2025年6月20日 上午10:30
四、中期答辩地点：自动化学院会议室410
五、中期答辩内容简介：
The research focuses on enhancing the sustainability and operational efficiency of cloud data centers through intelligent integration of renewable energy sources. As reliance on data centers grows, so do their energy demands and environmental impact. This work addresses those challenges through three key work packages.
In the first work package, a zonal-based tier-III data center architecture has been developed. The architecture divides the data center into thermal zones, enabling optimized workload distribution with thermal-aware job scheduling. This improves energy efficiency, ensures high uptime, reduces job queue time, and lowers carbon emissions through zonal battery storage and intelligent control.
The second work package completed a power demand prediction model based on thermal signatures, which captures internal zone behavior more accurately than traditional methods. A probabilistic renewable energy modeling approach was also introduced to assess the likelihood of generation meeting zonal power demand.
In the third work package, a dynamic energy mapping strategy was implemented using Mixed-Integer Linear Programming (MILP) to allocate distributed renewable sources across zones. Ongoing enhancements include a reinforcement learning-based control strategy for adaptive optimization and a climate-resilient planning framework incorporating IEEE bus systems, asset fragility modeling, and long-term infrastructure planning.
Together, these efforts support the development of a scalable, intelligent, and resilient data center energy management system that balances performance, cost, and environmental impact.
六、中期答辩人简介
Abubakar Rafique, PhD candidate in Electrical Engineering, School of Automation, Northwestern Polytechnical University. Research focuses on machine learning, deep learning, optimization and energy management.
