Efficient State of Health Estimation for LTO Batteries Using Liquid Time-Constant Neural Networks

In this paper, we present a novel approach for State of Health (SoH) estimation in Lithium Titanate Oxide (LTO) batteries using Liquid Time-Constant Neural Networks (LTCNNs). The LTCNN model is designed to capture complex temporal dynamics in capacity fading. We demonstrated LTCNN’s superior accuracy with a Root Mean Squared Error (RMSE) 0.01 on both training and test datasets. This high accuracy and the model’s low computational cost position LTCNN networks as a powerful real-time battery health monitoring tool. Our study highlights the LTCNN model’s compact architecture with a computational depth of 46 operations. Additionally, with only 131 parameters and an inference time of 0.51 s, while requiring 0.12 MB of memory allocation, LTCNNs achieve an efficient performance. These characteristics enable the LTCNN algorithm to be implemented on onboard chips such as Battery Management Systems (BMS). This feature allows continuous and real-time SoH estimation without needing high-powered computational resources. The results demonstrate that LTCNN neural networks offer a scalable and cost-effective solution for enhancing BMS performance in electric vehicles and other applications where LTO batteries are used. This paper contributes to the growing body of research on neural network-based SoH estimation, providing a practical framework for implementing LTCNN models in real-world systems.