Li-ion battery SOC estimation method using a Neural Network trained with data generated by a P2D model. Issue 10 (2021)

• Record Type: Journal Article
• Authors: Kuchly, Jean; Goussian, Alain; Merveillaut, Mathieu; Baghdadi, Issam; Franger, Sylvain; Nelson-Gruel, Dominique; Nouillant, Cédric; Chamaillard, Yann
• Is Part Of: IFAC-PapersOnLine. Volume 54:Issue 10(2021)
• Page Start: 336
• Page End: 343
• Publication Date: 2021
• Subjects: Lithium-ion Battery -- SOC estimation -- Neural Network -- Electrochemical Battery Model; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2021.10.185
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: The State Of Charge (SOC) estimation of a Li-ion battery is still an open problem. The most classical method, Coulomb Counting (CC) is vulnerable to current measurement bias. Measuring the Open-Circuit Voltage (OCV) allows to correct the error accumulated by the CC method, but only after the battery has been unsolicited long enough. Regarding these deficiencies, advanced SOC estimation methods try to combine current and voltage information, and are either based on an Extended Kalman Filter (EKF), which represents a certain algorithmic complexity and is hard to calibrate, or on black-box methods. In particular, methods using a Neural Network (NN) have been investigated in the literature, but take usually into account only instantaneous information, failing to represent the dynamic of ion diffusion in the electrodes. By considering also a close-past current integral as an input, this paper proposes a NN model able to correct initial SOC estimation errors and handle current measurement bias, and achieving a better estimation performance than a classical NN model taking only instantaneous information as an input.