Session: AVT-05-01 Advances in Vehicle Electrification and Powertrain Design

Paper Number: 89454

89454 - Assessment of State of Charge Estimation Methods Based on Neural Networks and Support Vector Machine for Lithium-Ion Batteries Used in Vehicular Applications 

The State of Charge (SOC) estimation in Lithium-ion batteries is a challenging task that is currently assessed with different methods in a vast variety of applications. This paper presents the design and assessment of two SOC estimation methods, based on Artificial Neural Networks (ANNs) and Support Vector Machine (SVM) algorithms for Lithium-ion batteries used in vehicular applications. The paper validates the two proposed approaches with experimental data collected during a laboratory test campaign. The obtained results are compared in terms of estimation accuracy, proving the feasibility of the considered algorithms. Moreover, the paper describes the retained software architectures and the design procedure related to the two proposed techniques based on Artificial Intelligence (AI). In detail, the retained Lithium-ion battery is a 21.6V 3.3Ah battery pack that is used as an energy module for vehicular applications. The considered battery module is numerically modeled with a second-order RC equivalent Thevenin model to collect a sufficient amount of data for the algorithms' design phase. The model parameters are identified with a grey-box approach based on a non-linear least squares algorithm. The designed algorithms can accurately estimate the battery SOC with both the ANN-based and SVM-based methods. Specifically, the resulting mean prediction error is always below 2.5% and 3.5% for the ANN-based and SVM-based algorithms, respectively.

Presenting Author: Sara Luciani Politecnico di Torino

Presenting Author Biography: NA

Authors:

Sara Luciani Politecnico di Torino
Stefano Feraco Politecnico di Torino
Mario Silvagni Politecnico di Torino
Angelo Bonfitto Politecnico di Torino
Nicola Amati Politecnico di Torino
Andrea Tonoli Politecnico di Torino