Data-Driven State of Charge Estimation for Lithium-Ion Batteries Based on Polynomial Surface Fitting Under Dynamic Driving Cycles

Abstract

Accurate state-of-charge (SOC) estimation is essential for battery management systems (BMSs) under dynamic operating conditions. This study proposes a lightweight data-driven SOC estimation framework based on a tensor-product bivariate polynomial surface. During offline model identification, the reference SOC and measured current are used to establish an explicit terminal-voltage surface (V = f(I, SOC)). During SOC estimation, however, the reference SOC is no longer used as an input; instead, SOC is directly recovered from the measured terminal voltage and current by inversion of the identified polynomial surface. The method used in this study is developed and evaluated using three public dynamic driving-cycle datasets, including DST, FUDS, and UDDS. To reduce temporal information leakage, the datasets are partitioned chronologically, and the polynomial orders are selected using five-fold forward-chaining cross-validation. The selected third-order current and fifth-order SOC polynomials provide a compact 24-coefficient representation. Direct comparison between the estimated and reference SOC yields an RMSE of 2.91% on the DST test set, while the corresponding RMSE values on FUDS and UDDS are 3.18% and 3.54%, respectively. The results demonstrate that the proposed approach provides a computationally lightweight and interpretable SOC estimation framework without requiring recursive filtering or online equivalent-circuit parameter identification.