A novel enhanced SOC estimation method for lithium-ion battery cells using cluster-based LSTM models and centroid proximity selection

In line with the global mission in achieving the net zero target through deployment of renewable energy technologies and electrifying the transportation sector; precise and adaptable State of Charge (SOC) estimation for Lithium-ion batteries has emerged as a critical need. The paper introduces a novel Cluster-Based Learning Model (CBLM) framework that integrates the strengths of K-Means and Fuzzy C-Means clustering with the predictive power of Long Short-Term Memory (LSTM) networks. This approach aims to enhance the precision and reliability of battery SOC estimations, adapting to the dynamic and complex operational conditions characteristic of Li-ion batteries. The key contributions of this study is the development and validation of the CBLM framework, which was proven to outperform state-of-art standalone deep learning techniques particularly under diverse operational conditions. Additionally, the introduction of a centroid proximity selection mechanism within the CBLM framework, which dynamically selects the most appropriate cluster model in real-time based on the proximity of the operational data to the cluster centroids. The performance of the proposed CBLM approach is evaluated using a Tesla Model 3 2170 Li-ion battery dataset. Results demonstrate the model's enhanced performance, with reductions in Root Mean Square Error (RMSE) to as low as 0.65% and Mean Absolute Error (MAE) to 0.51%, reducing state-of-art benchmark model errors by margins of 61.8% and 68.5% respectively. Additionally, the maximum error using CBLM was lower than benchmark, emphasising the model's reliability in worst-case-scenarios. The study also conducted comprehensive ablation tests on the proposed novel framework to further optimise its performance.