Design and Simulation of an Electrochemical ANN-Based Observer for a Lithium-Ion Battery for Estimation of State of Charge

Finite Element Methods (FEM) for estimation of State Of Charge (SOC) of the Lithium-ion battery are inaccurate. A more accurate and faster method using Artificial Neural Networks (ANNs) is investigated in this research. The SOC of the Lithium-ion battery was determined from an electrochemical mathematical model of the Lithium-ion battery drawn from current literature. The model turns out to be complex as it is described in terms of non-linear Partial Differential Equations (PDE). Since Luenberger observer design is not applicable to systems described by linear or non-linear PDEs, an observer was designed based on back-stepping approach to estimate SOC of the Lithium-ion battery. The ANNs are implemented and then trained using MATLAB Neural Network Toolbox. The results for undertaking this research are a more accurate estimation of SOC of a Lithium ion battery for real-time applications e. g. in a dual source Pure Electric Vehicle (PEV). The proposed ANN-based design and simulation of the PDE observer has been validated by comparing the error of the observer based on ANNs and the same observer based on FEM. Simulations of the ANN-based observer and FEM-based observer were performed in MATLAB.