Accurate residual capacity estimation of retired LiFePO4 batteries based on mechanism and data-driven model. (1st January 2022)

• Record Type: Journal Article
• Authors: Ni, Yulong; Xu, Jianing; Zhu, Chunbo; Pei, Lei
• Is Part Of: Applied energy. Volume 305(2022)
• Publication Date: 2022-01-01
• Subjects: Second-use applications -- Retired LiFePO4 batteries -- Residual capacity estimation -- Mechanism model -- Data-driven model; ANN artificial neural networks -- BMS battery management system -- CC constant current -- CCS charging curve sections -- CNN convolutional neural network -- CV constant voltage -- DAAM discrete Arrhenius aging model -- EV electric vehicle -- FNN feedforward neural network -- GPR Gaussian process regression -- HIF H infinity filter -- HIs health indicators -- IMFO improved moth-flame optimization -- KF Kalman filter -- LAM loss of active material -- LLI loss of lithium inventory -- LSTM long short-term memory -- MAE mean absolute error -- MFO moth-flame optimization -- MVF mean voltage falloff -- PF particle filter -- PMM prognostic and mechanistic -- RBF radial basis function -- RLS recursive least squares -- RMSE root mean square error -- SEI solid electrolyte interphase -- SOC state of charge -- SOH state of health -- SVR support vector regression; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2021.117922
• Languages: English
• ISSNs: 0306-2619

Abstract

Highlights: A hybrid method for residual capacity estimation is proposed. Three new health indicators related to the capacity loss mechanism are established. Support vector regression by improved moth-flame optimization is proposed. Residual capacity tests of 1000 retired batteries are conducted. The root mean square error is within 2.18% using only the first 10% of the data. Abstract: Accurate residual capacity estimation of retired LiFePO4 batteries is critically important for second-use applications but is challenging with multiple aging pathways and nonlinear degradation mechanisms. In this study, a fast and accurate residual capacity estimation method based on the mechanism and data-driven model is developed with two main contributions. First, as the basis of the residual capacity estimation model, three new health indicators directly related to the capacity loss mechanism are derived from the prognostic and mechanism model using the Levenberg-Marquardt method and Spearman correlation. Second, residual capacity tests were conducted on 1000 retired batteries to establish a data-driven model for residual capacity estimation based on the proposed health indicators, guaranteeing better universality and estimation accuracy for different types of retired LiFePO4 batteries. To establish a data-driven model for the residual capacity estimation, an improved moth–flame optimization and support vector regression method is used; the adaptive weight and Levy flight are introduced in the moth–flame optimization algorithm to prevent the local optimal value. The residual capacity estimation results are compared with the results from three other typical methods and input health indicators. The results show that the root mean square error of the proposed method is within 2.18% using only the first 10% of the data, a smaller error than with the other methods. A fast and accurate residual capacity estimation method for retired batteries can reduce the cost and improve the development for second-use applications.