To ensure the safety and functionality, it is crucial to evaluate the state of battery system in electric vehicles. Describing battery dynamic characteristic with a combined state space model, this paper presents a novel adaptive nonlinear neural observer based approach for state of charge estimation, which is composed of linear discriminant function, nonlinear neural proportional-integral observer and extended Kalman filter. It incorporates the local linear approximation capability of extended Kalman filter with the nonlinear mapping, self-learning and self-adjusting capabilities of neural proportional-integral observer, which is used to compensate the deviation resulted from the underestimated initial state, process noise and measurement noise. Taking the samples collected from lithium-ion battery test system for example, simulation is carried out to verify the proposed method. Results show that it is capable of evaluating the state of charge of cell with a rapid convergence and an error <2 % while remaining unaffected by the unknown initial cell states and the underestimation of the process noise and the measurement noise.
Read full abstract