Soc Estimation Using a Neural Network Trained with Model-Based Dataset Implemented in Ground Electric Vehicles BMS

The battery pack is a key subsystem in electric vehicles and requires on-board intelligent management devices. Among monitoring and diagnostic operations, estimating State-of Charge (SoC) is one of the most important tasks to be addressed by the Battery Management System (BMS). Data-driven methods have a great potential to estimate SoC under different environmental and load conditions. However, they require the collection of large training datasets, which is a time and resourceconsuming task. To address this challenge, this paper demonstrates the validity of using a model-based approach to easily generate training data for Feed Forward Neural Networks (FFNNs) SoC estimators, which is validated with a reduced number of dedicated experimental tests reflecting real driving operative conditions. The proposed methodology is applied to the case study of a Li-NMC storage cell. The trained NN has been successfully implemented on a microcontroller to prove its realtime applicability in BMS boards. The proposed methodology can be leveraged to scale the estimator to the cases of batteries with different technologies, sizes and applications.