A hybrid statistical data-driven method for on-line joint state estimation of lithium-ion batteries

• Published in: Applied energy 2020-03, Vol.261, p.114408, Article 114408
• Main Authors: Song, Yuchen, Liu, Datong, Liao, Haitao, Peng, Yu
• Format: Article
• Language: English
• Subjects: Joint state estimation; Lithium-ion battery; State-of-charge; State-of-health; Statistical data-driven model fusion
• ISSN: 0306-2619; 1872-9118
• DOI: 10.1016/j.apenergy.2019.114408

•A lithium-ion battery on-line joint state estimation model is developed.•State-of-charge and state-of-health are estimated with a fused statistical model.•Battery degradation is characterized by terminal voltage and electrical current.•State-of-health is the feedback of state-of-charge estimates for higher accuracy.•The method is capable to be used in complex practical applications. Lithium-ion batteries are increasingly being used as the energy storage systems in electric vehicles, smart grid and aerospace systems. Estimating the state-of-charge and state-of-health of lithium-ion battery is essential to the operational safety and reliability of a system. However, some useful battery parameters, such as capacity and impedance, are not easy to measure because of the complex testing procedure and conditions. This makes the on-line state-of-health estimation suffer from low accuracy and further impacts the estimation accuracy of state-of-charge. This paper proposes a joint lithium-ion battery state estimation approach that takes advantage of the data-driven least-square-support-vector-machine and model-based unscented-particle-filter. The indicator of battery performance degradation is extracted for state-of-health estimation based on the measurable terminal voltage and electric current. Then, the least-square-support-vector-machine is implemented to provide direct and nonlinear mapping models for state-of-health and state-of-charge. These direct mapping models avoid parameter identification and updating, which are applicable in cases with complex operating conditions. Unscented-particle-filter is utilized to take the least-square-support-vector-machine estimates as the temporal measurements for optimal state-of-health and state-of-charge estimation. The state-of-health correction in state-of-charge estimation achieves the joint estimation with different time scales. An experimental study on battery dynamic stress tests illustrates that the life cycle maximum state-of-charge estimation error is less than 2% and the root-mean-square-error of state-of-health estimation is less than 4%, which mean both state-of-charge and state-of-health can be estimated with high accuracy and robustness using the proposed hybrid joint state estimation method.