State-of-Health Estimation for Lithium-Ion Batteries Based on the Multi-Island Genetic Algorithm and the Gaussian Process Regression

• Published in: IEEE access Vol. 5; pp. 21286 - 21295
• Main Authors: Wang, Zhenpo, Ma, Jun, Zhang, Lei
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.01.2017; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Batteries; Conjugate gradient method; Correlation analysis; Cost analysis; Degradation; Electric vehicles; Estimation; Experimentation; Gaussian process; Gaussian process regression; Gaussian processes; Genetic algorithms; incremental capacity analysis; Integrated circuits; Kernel functions; Lithium; Lithium-ion batteries; Management systems; Mathematical models; Model accuracy; multi-island genetic algorithm; Optimization; Parameter estimation; Parameter identification; Power management; Rechargeable batteries; Regression analysis; state of health; Wavelet transforms
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2017.2759094

Battery State-of-Health (SOH) estimation is of utmost importance for the performance and cost-effectiveness of electric vehicles. Incremental capacity analysis (ICA) has been ubiquitously used for battery SOH estimation. However, challenges remain with regard to the characteristic parameter selection, estimation viability and feasibility for practical implementation. In this paper, a novel ICA-based method for battery SOH estimation is proposed, with the goals to identify the most effective characteristic parameters of IC curves, optimize the SOH model parameters for better prediction accuracy and enhance its applicability in realistic battery management systems. To this end, the IC curve is first derived and filtered using the wavelet filtering, with the peak value and position extracted as health factors (HFs). Then, the correlations between SOH and HFs are explored through the grey correlation analysis. The SOH model is further established based on the Gaussian process regression (GPR), in which the optimal hyper parameters are calculated through the conjugate gradient method and the multi-island genetic algorithm (MIGA). The effects of different HFs and kernel functions are also analyzed. The effectiveness of the proposed MIGA-GPR SOH model is validated by experimentation.