Active cell equalization for lithium-ion battery packs in electric vehicles using state of power estimation with convolutional neural network

• Published in: Energy conversion and management. X Vol. 27; p. 101100
• Main Authors: Shreasth, Ooi, Chia Ai, Khan, Neha, Desa, Mohd Khairunaz Bin Mat, Amir, Mohammad, Ishak, Mohamad Khairi
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.07.2025; Elsevier
• Subjects: Active cell equalization; Convolution neural network; Electric vehicles; Lithium-ion battery pack; State-of-Charge; State-of-Power; Lithium-ion battery pack; Electric vehicles; Active cell equalization; Convolution neural network; State-of-Charge; State-of-Power
• ISSN: 2590-1745; 2590-1745
• DOI: 10.1016/j.ecmx.2025.101100

•Introduced a State-of-Power (SoP) cell equalization algorithm to ensure optimal distribution of charge among cells in the battery pack.•Developed a U-net Convolution Neural Network (CNN) model for accurate SoP estimation under varying load current profiles.•Designed a Cell-to-Pack-to-Cell (CTPTC) equalization circuit in MATLAB/Simulink to compare State-of-Charge (SoC) and SoP equalization strategies.•Built an experimental setup to validate the effectiveness and superiority of the proposed SoP algorithm. In Electric Vehicles (EVs), the battery pack is composed of hundreds or even thousands of Lithium-ion (Li-ion) cells connected in series to deliver the required power and energy for vehicle operation. However, the charge imbalance among series connected cells is inevitable due to inconsistent manufacturing processes and environmental conditions. To address this issue, cell equalization is essential to balance the charge distribution among Li-ion cells within the battery pack. This paper proposes an active cell equalization algorithm based on State-of-Power (SoP), which outperforms voltage and State-of-Charge (SoC) based equalization by reducing balancing losses and increasing the usable capacity of the battery pack. To achieve accurate SoP estimation, it is essential to accurately determine the SoP of each cell. Therefore, a new SoP estimation method utilizing a Convolution Neural Network (CNN) with U-net architecture is proposed. This approach achieves precise SoP prediction, with a Root Mean Square Error (RMSE) of less than 0.138 under Urban Dynamometer Driving Schedule (UDDS) drive cycle conditions. The proposed SoP based cell equalization algorithm is validated through simulations on the MATLAB/Simulink platform, demonstrating its ability to converge the SoC and SoP of individual cells, ensuring balanced charge distribution with minimal balancing efforts. Furthermore, a hardware experiment using 24Li-ion cells is conducted to confirm its practicality and reliability. Compared to the SoC based algorithm, the proposed SoP method increases usable capacity by 1.8%, enhancing battery pack longevity, safety, and overall efficiency, making it an ideal solution for EV battery systems.