Breakthrough in battery management: New AI model enables more accurate state of charge estimation

Researchers from Hebei University of Technology have developed an innovative approach to lithium-ion battery management that could significantly improve the safety and performance of electric vehicles. The new method, which estimates a battery's state of charge (SOC) based on its internal core temperature rather than surface temperature, addresses a critical flaw in current battery management systems while requiring less computational resources. SOC estimation—essentially determining how much "fuel" remains in an electric vehicle's battery—is crucial for safe and efficient battery operation. However, current methods typically rely on surface temperature measurements that can be misleading for the accuracy of SOC estimation.

"The difference between surface and core temperatures in lithium-ion batteries can reach 5-10°C under severe conditions," explains the research team. "Using surface temperature as an input feature for SOC estimation is fundamentally unreasonable and can lead to significant errors." This temperature discrepancy is particularly problematic during fast charging or high-power operations, precisely when accurate battery management is most critical for safety and performance.

The researchers tackled this challenge with a two-step approach. First, they developed a method to estimate the battery's core temperature—what they call the State of Temperature (SOT)—using only externally measurable parameters. Then, they used this estimated core temperature as one of the input parameters for more accurate SOC estimation. For both steps, they employed a novel neural network architecture called Weight Clustered-Convolutional Neural Network-Long Short-Term Memory (WC-CNN-LSTM), which maintains high accuracy while dramatically reducing the model's size. One of the most significant innovations in this research is the development of the WC-CNN-LSTM model, which addresses a major limitation of deep learning approaches in battery management systems. This breakthrough makes it feasible to implement sophisticated battery management algorithms directly in the resource-constrained embedded systems used in electric vehicles, rather than requiring powerful external computing resources.

By more accurately estimating a battery's true state of charge, especially in extreme conditions, the system can help prevent dangerous situations like over-charging or over-discharging. It also provides drivers with more reliable information about their vehicle's remaining range, potentially reducing range anxiety—a major barrier to EV adoption. Additionally, the compact size of the model makes it practical to implement in current vehicle computer systems without requiring hardware upgrades.

While the current research demonstrates significant improvements in both accuracy and efficiency, the team suggests several directions for future work:

1. Extending the model to account for battery aging effects, which can further complicate temperature and charge estimation

2. Implementing the system in real-world electric vehicles to validate performance under diverse driving conditions

3. Exploring additional model compression techniques to further reduce computational requirements

As electric vehicles continue to gain market share worldwide, innovations like this temperature-based SOC estimation approach will play a crucial role in making these vehicles safer, more reliable, and more appealing to consumers still hesitant about making the switch from conventional vehicles.

Reference

Author: Chaoran Li ^a b, Sichen Zhu ^a b, Liuli Zhang ^d, Xinjian Liu ^a b, Menghan Li ^a b, Haiqin Zhou ^e, Qiang Zhang ^c, Zhonghao Rao ^a b

Title of original paper: State of charge estimation of lithium-ion battery based on state of temperature estimation using weight clustered-convolutional neural network-long short-term memory

Article link: https://doi.org/10.1016/j.geits.2024.100226

Journal: Green Energy and Intelligent Transportation

https://www.sciencedirect.com/science/article/pii/S2773153724000781

DOI: 10.1016/j.geits.2024.100226

Affiliations:

^a Hebei Engineering Research Center of Advanced Energy Storage Technology and Equipment, School of Energy and Environmental Engineering, Hebei University of Technology, Tianjin 300401, China

^b Hebei Key Laboratory of Thermal Science and Energy Clean Utilization, School of Energy and Environmental Engineering, Hebei University of Technology, Tianjin 300401, China

^c School of Energy and Power Engineering, Shandong University, No. 17923 Jingshi Road, Lixia District, Jinan 250061, China

^d Pinggao Group Energy Storage Technology Co., Ltd., Room 1-4251, Block E, No. 6 Huafeng Road, Huaming High-tech Industrial Zone, Dongli District, Tianjin 300308, China

^e Chemical and Environmental Engineering, University of Park, Nottingham, NG7 2RD, UK