A method for state of energy estimation of lithium-ion batteries based on neural network model. (October 2015)

• Record Type: Journal Article
• Authors: Dong, Guangzhong; Zhang, Xu; Zhang, Chenbin; Chen, Zonghai
• Is Part Of: Energy. Volume 90:Part 1(2015)
• Page Start: 879
• Page End: 888
• Publication Date: 2015-10
• Subjects: State of energy -- Wavelet neural network -- Lithium-ion batteries -- Particle filter; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2015.07.120
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: The state-of-energy is an important evaluation index for energy optimization and management of power battery systems in electric vehicles. Unlike the state-of-charge which represents the residual energy of the battery in traditional applications, state-of-energy is integral result of battery power, which is the product of current and terminal voltage. On the other hand, like state-of-charge, the state-of-energy has an effect on terminal voltage. Therefore, it is hard to solve the nonlinear problems between state-of-energy and terminal voltage, which will complicate the estimation of a battery's state-of-energy. To address this issue, a method based on wavelet-neural-network-based battery model and particle filter estimator is presented for the state-of-energy estimation. The wavelet-neural-network based battery model is used to simulate the entire dynamic electrical characteristics of batteries. The temperature and discharge rate are also taken into account to improve model accuracy. Besides, in order to suppress the measurement noises of current and voltage, a particle filter estimator is applied to estimate cell state-of-energy. Experimental results on LiFePO4 batteries indicate that the wavelet-neural-network based battery model simulates battery dynamics robustly with high accuracy and the estimation value based on the particle filter estimator converges to the real state-of-energy within an error of ±4%. Highlights: State-of-charge is replaced by state-of-energy to determine cells residual energy. The battery state-space model is established based on a neural network. Temperature and current influence are considered to improve the model accuracy. The particle filter is used for state-of-energy estimation to improve accuracy. The robustness of new method is validated under dynamic experimental conditions.