The state of charge (SoC) of a battery, when inaccurately estimated can lead to customer dissatisfaction and major inconvenience especially in context of electric vehicles (EVs). In this paper, the hardware implementation of an algorithm for SoC accurate estimation, based on gaussian process regression (GPR), surrogate based optimized long short-term memory (SO-LSTM) neural network (NN) and custom digital signal processing is presented. The implemented algorithm was trained and validated with constant current discharge profiles acquired from a Panasonic 18650 LithiumIon battery (LiB). A current range of 2-3 A was selected for the hardware implementation. The proposed machine learning (ML) model was designed in MATLAB/Simulink environment, then the code was automatically generated for the deployment on the STM32F411RE nucleo board. The average RMSE of the hardware deployed model results in 0.74% and the inference time related to a single sample was 1.108 ms. These results are fully compatible with implementation in a real scenario and show better performance when compared with literature.
Ali, S., Bianchi, V., De Munari, I. (2025). A Microcontroller Based Optimized Framework for the State of Charge Estimation of a Lithium Ion Battery. 345 E 47TH ST, NEW YORK, NY 10017 USA : IEEE [10.1109/MetroAutomotive64646.2025.11119267].
A Microcontroller Based Optimized Framework for the State of Charge Estimation of a Lithium Ion Battery
Ali S.
;
2025
Abstract
The state of charge (SoC) of a battery, when inaccurately estimated can lead to customer dissatisfaction and major inconvenience especially in context of electric vehicles (EVs). In this paper, the hardware implementation of an algorithm for SoC accurate estimation, based on gaussian process regression (GPR), surrogate based optimized long short-term memory (SO-LSTM) neural network (NN) and custom digital signal processing is presented. The implemented algorithm was trained and validated with constant current discharge profiles acquired from a Panasonic 18650 LithiumIon battery (LiB). A current range of 2-3 A was selected for the hardware implementation. The proposed machine learning (ML) model was designed in MATLAB/Simulink environment, then the code was automatically generated for the deployment on the STM32F411RE nucleo board. The average RMSE of the hardware deployed model results in 0.74% and the inference time related to a single sample was 1.108 ms. These results are fully compatible with implementation in a real scenario and show better performance when compared with literature.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
