Optimized multi-hidden layer long short-term memory modeling and suboptimal fading extended Kalman filtering strategies for the synthetic state of charge estimation of lithium-ion batteries. (15th April 2023)

• Record Type: Journal Article
• Authors: Xie, Yanxin; Wang, Shunli; Zhang, Gexiang; Fan, Yongcun; Fernandez, Carlos; Blaabjerg, Frede
• Is Part Of: Applied energy. Volume 336(2023)
• Publication Date: 2023-04-15
• Subjects: Ternary lithium-ion battery -- Long time series -- Long short-term memory network -- Hyper-parameter selection -- Suboptimal fading factor extended Kalman filtering algorithm -- Custom driving conditions; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2023.120866
• Languages: English
• ISSNs: 0306-2619

Abstract

Abstract: With the demand for high-endurance lithium-ion batteries in new energy vehicles, communication and portable devices, high energy density lithium-ion batteries have become the main research direction of the battery industry. State of Charge (SoC), as a state parameter that must be accurately evaluated by the battery management system, enables online safety monitoring of the battery operation, and prolongs its service life. In this paper, an improved algorithm based on multi-hidden layer long short-term memory (MHLSTM) neural network and suboptimal fading extended Kalman filtering (SFEKF) is proposed for synthetic SoC estimation. First, the battery external measurable information is captured. The battery real data properties are matched with the network topology without additional battery model construction, and the battery SoC is roughly evaluated using an MHLSTM network. Then, a suboptimal fading factor is inserted into the extended Kalman filter (EKF) algorithm for iterative recursion and adaptive handling to smooth the prediction results of the MHLSTM network and enhance the accuracy of state estimation, system stability, and generality. Three customized electric vehicle (EV) driving conditions datasets are categorized into training and testing sets to fulfill the efficient estimation of synthetic SoC by the fusion algorithm and solve the time series problem. Using the maximum error (ME), mean absolute error (MAE), root mean squared error (RMSE), and mean absolute percentage error (MAPE), the results show that the maximum bias of the fusion algorithm to estimate the synthetic SoC is limited to within 1.2%, even under the abrupt change of the system. It can converge to the real value quickly and maintains an excellent tracking capability for data changes, reflecting the high accuracy estimation capability and the robustness possessed by the system. Highlights: A fusion model is presented for synthetic SoC estimation. The MHLSTM network topology is matched with the actual battery measurement data. The SFEKF algorithm is developed to adaptively smooth the MHLSTM model results. The accuracy of the model is verified by simulation tests on different data sets. The method features excellent generalization ability and can improve the estimation accuracy.