A state-dependent quasi-linear parameter-varying model of lithium-ion batteries for state of charge estimation

Abstract

Accurate estimation of state of charge (SOC) forms the foundation of battery management systems. Although commonly used for SOC estimation, equivalent circuit models (ECMs) inadequately capture battery nonlinear dynamics and rely on SOC-open circuit voltage curves. To overcome these limitations, this paper introduces state-dependent mechanisms into ECMs, proposing a state-dependent quasi-linear parameter-varying model (SD-QLPVM). This model incorporates a quasi-linear model derived from ECMs. Crucially, it eschews traditional approaches of parameter determination through offline experiments or online adaptive methods, which are limited by their linear nature. Conversely, the parameters of the quasi-linear model are treated as time-varying and state-dependent functional parameters, calculated using radial basis function neural networks (RBF-NNs). Subsequently, state variables, such as terminal voltage, current, SOC, and temperature, are used to characterize the operation point of LIBs. By considering state variables as the inputs to the RBF-NNs, the proposed parameter determination approach enables the quasi-linear model to dynamically adjust its parameters in response to evolving battery operation points, representing battery dynamics accurately and responsively. Finally, an online SOC estimation method is developed based on the SD-QLPVM and a particle filter. The effectiveness of the proposed model and SOC estimation method is verified across various drive cycles and temperature conditions.