Adaptive Neural Network-Based Prescribed-Time Observer for Battery State-of-Charge Estimation

Convergence speed is an important indicator to evaluate the performance of state-of-charge (SOC) estimators. To improve the convergence speed, this article proposes an adaptive radial basis function neural network-based prescribed-time observer to estimate the battery SOC. First, an adaptive RBF NN is employed to approximate the nonlinear part of the battery equivalent circuit model, and the online learning process of network weight can adapt the variations in battery parameters. Then, a prescribed-time SOC observer is developed to ensure the state and weight estimation errors converge within the convergence time
$T$
, which can be prescribed by users and is irrelevant on initial values. Thus, the network weight no longer needs to update when time exceeds
$T$
, and the computational burden can be effectively saved. Furthermore, a switched-gain scheme with a naturally switched time
$T$
is employed to simultaneously guarantee the convergence speed and estimation accuracy. An adaptive robust term is designed to compensate the approximation error and possible variations of the network weight in the steady state. Finally, the theoretical stability is proved by the Lyapunov theory, and the practical effectiveness is evaluated by experiments and simulations.