Online Learning Approach for Predictive Real-Time Energy Trading in Cloud-RANs

Area of Science:

• Telecommunications Engineering
• Energy Systems Management
• Machine Learning Applications

Background:

• Electricity demand variability and uncertainty impact both power generation and cellular communication systems.
• Cloud Radio Access Networks (Cloud-RANs) face challenges in maintaining cost-aware and reliable operations due to these uncertainties.

Purpose of the Study:

• To develop an online learning algorithm for prescheduling energy procurement in Cloud-RANs.
• To manage variability and uncertainty for cost-efficient and reliable network operation.

Main Methods:

• Development of a combinatorial multi-armed bandit model for energy scheduling.
• Implementation of an algorithm that learns from past cooperative energy trading and explores new scheduling strategies.
• Minimization of long-term energy costs at remote radio heads.

Main Results:

• The proposed algorithm effectively controls available power budgets.
• Significant reduction in overall energy costs demonstrated through simulations.
• Outperforms recently proposed approaches for real-time energy resources and trading in Cloud-RANs.

Conclusions:

• The developed online learning algorithm offers a robust solution for energy management in Cloud-RANs.
• The approach successfully balances cost minimization with reliable network operation.
• Enables efficient utilization of ancillary energy markets for future time slots.