Robust state of health estimation and remaining useful life prediction for lithium-ion battery with synchronous framework using data-driven integrated model

Accurate state of health estimation and remaining useful life prediction can improve the safety and prolong the service life of lithium-ion battery system. This paper proposes a robust framework for state of health estimation and remaining useful life prediction based on data-driven integrated model. Firstly, three representative health indicators are extracted to reflect the aging state of the battery, and these health indicators are developed based on the energy information in charging and discharging process. Secondly, a data-driven integrated model of battery ageing state-space representation is developed, in which Gaussian process regression is employed to establish state equation using historical capacity series and current capacity, and maps the relationship between capacity degradation and health indicators to construct an observation equation. Thirdly, the particle filter is introduced to realize the closed-loop estimation of battery capacity and suppress the measurement noises by combining with the data-driven integrated model and regarding current extracted health features as observations. Meanwhile, available estimated capacity is fed back to the model to build the dynamic architecture. Fourthly, for the unavailability of observations in remaining useful life prediction issue, an autoregressive model is introduced to roll predict the future observation from the historical health features to complete the closed-loop synchronous framework, and an error compensation mechanism based on the extreme learning machine scheme is proposed to further enhance the accuracy of estimation. Finally, a practical aging experience involving 7 batteries are performed, and the experimental results illustrate that the proposed method can guarantee high accuracy and robustness relatively.