Loading...

Optimal Energy Management in Battery Systems with an Intelligent Approach

Tavakol Moghaddam, Yasaman | 2025

30 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 58691 (46)
  4. University: Sharif University of Technology
  5. Department: Energy Engineering
  6. Advisor(s): Boroushaki, Mehrdad
  7. Abstract:
  8. Cell balancing in lithium-ion battery packs plays a critical role in enhancing safety, improving performance, and extending battery system lifespan. This dissertation presents a novel and intelligent framework for optimal battery cell balancing, based on reinforcement learning (RL), and covering both dissipative and non-dissipative approaches. Firstly, the first application of reinforcement learning to dissipative (passive) balancing is introduced. A deep reinforcement learning agent is designed using the Trust Region Policy Optimization (TRPO) algorithm. The agent learns a sophisticated management policy that increases the usable capacity of the pack by up to 16.8%, reduces the state-of-charge variance by up to 69.4%, and lowers the number of switching operations by 40.4%. Without the need for retraining, the learned policy generalizes well and performs successfully across diverse battery packs with different configurations. In the second part, we propose the first application of multi-agent reinforcement learning (MARL) to any type of battery cell balancing, specifically targeting non-dissipative (active) balancing suitable for electric vehicles. In this decentralized system, each battery cell acts as an autonomous agent that collaborates with others via a low-voltage bus. Under load profiles derived from real-world driving data, this approach increases usable capacity by up to 28% and achieves balancing up to three times faster than rule-based methods. The key contributions of this research include the first-ever use of RL for dissipative balancing, the first use of MARL for battery cell balancing of any kind, the design of generalizable learning algorithms, multi-objective reward functions, and the emergence of optimized management policies. These innovations pave the way toward the next generation of intelligent battery management systems, offering a validated roadmap for deploying smarter, safer, and longer-lasting energy storage solutions in industry
  9. Keywords:
  10. Lithium Ion Batteries ; Battery Cell Balancing ; Reinforcement Learning ; Battery Management System ; Multi-Agent Reinforcement Leaning ; Optimal Energy Management ; Smart Energy Management

 Digital Object List

 Bookmark

...see more