Privacy-Preserving Energy Management of a Shared Energy Storage System for Smart Buildings: A Federated Deep...

Area of Science:

• Smart Grid Technology
• Artificial Intelligence
• Energy Systems

Background:

• Smart buildings require efficient energy management to optimize consumption and integrate with shared resources.
• Privacy concerns hinder the adoption of centralized energy management systems for multiple buildings.
• Shared energy storage systems (SESS) offer potential for cost savings and grid stability but require sophisticated control.

Purpose of the Study:

• To propose a privacy-preserving energy management framework for a shared energy storage system (SESS) serving multiple smart buildings.
• To leverage federated reinforcement learning (FRL) to enable decentralized learning while protecting sensitive building energy consumption data.
• To optimize the charging and discharging of the SESS and the energy consumption of building systems.

Main Methods:

• A distributed deep reinforcement learning (DRL) framework utilizing federated reinforcement learning (FRL) with a global server (GS) and local building energy management systems (LBEMSs).
• LBEMS DRL agents share anonymized neural network components (energy consumption models) with the GS, excluding raw consumption data.
• The GS constructs a global energy consumption model for LBEMS retraining and trains a SESS agent for optimal charging/discharging operations.

Main Results:

• The proposed FRL approach effectively schedules SESS charging and discharging operations.
• Optimal energy consumption for heating, ventilation, and air conditioning (HVAC) systems in smart buildings was achieved.
• The framework successfully preserved the privacy of individual buildings' energy consumption data.

Conclusions:

• Federated reinforcement learning provides a viable solution for privacy-preserving energy management in multi-building SESS.
• The developed distributed DRL framework enables efficient energy scheduling and consumption optimization without compromising user privacy.
• The approach demonstrates robustness in heterogeneous building environments, paving the way for secure smart grid integration.