Premium
A novel framework for lithium‐ion battery state of charge estimation based on Kalman filter Gaussian process regression
Author(s) -
Chen Xiaobing,
Chen Xiaoping,
Chen Xiwen
Publication year - 2021
Publication title -
international journal of energy research
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.808
H-Index - 95
eISSN - 1099-114X
pISSN - 0363-907X
DOI - 10.1002/er.6649
Subject(s) - state of charge , kalman filter , battery (electricity) , state of health , covariance , control theory (sociology) , gaussian process , computer science , extended kalman filter , parametric statistics , kriging , engineering , process (computing) , gaussian , artificial intelligence , machine learning , mathematics , statistics , power (physics) , physics , control (management) , operating system , quantum mechanics
Summary Modeling and knowing states of lithium batteries have critical tasks, which are vital to understand battery behaviors and to guarantee the reliability and safety of battery energy storage system. Accurate prediction state of charge can provide important parameters to minimize the maintenance costs and prolong the battery life through optimal maintenance scheduling. However, the complex physicochemical features of the battery lead to the nonlinear characteristics of battery, which increases the difficulties of establishing battery model. Herein, a novel non‐parametric model for lithium battery based on the Gaussian process regression is proposed. First, the Gaussian process regression is applied to build the process and observation models through learning the experimental battery datasets offline. The unscented Kalman filter algorithm is then utilized to improve the accuracy of state of charge estimation through correcting the voltage errors. Otherwise, the mean and the covariance function of the prediction model are employed to indicate the uncertainty model. Two types of batteries are employed for demonstrating the capability and efficacy of the proposed method under various driving cycles, temperatures, and health levels. The maximum relative error is within 2% for these three scenarios indicate the proposed method can provide accurate and robust state of charge estimation.