As lithium-ion batteries are the main power source of new energy vehicles, making accurate predictions of unknown State of Charge (SOC) during vehicle operation for vehicle data monitoring is vital to the advancement of intelligent new energy vehicles. In this manuscript, an expression tree-based genetic programming regression model (ETGPR) is proposed to estimate the real-time SOC of lithium-ion batteries. The proposed model mainly adopts the symbolic regression technique. In addition to the current–voltage curves being fed into the model, an additional approach is designed to ensure real-time model predictions in dynamic situations, which includes the previous moment's power in the input parameters. Different seed hyperparameters in the model are set, and the model automatically performs evolutionary calculations. Subsequently, each parameter of the model is optimally adjusted to obtain a set of regression expressions that accurately reflect the relationship between the SOC and each parameter after a specified number of iterations. Finally, the generated expression is proven to perform better in terms of its ability to capture the nonlinear relationship between SOC and battery variables. Also, the model demonstrates excellent robustness in the presence of notable noise from input-independent features compared to other models, a root mean square error (RMSE) of less than 0.3% and a mean absolute error (MAE) of less than 0.2% are achieved. Furthermore, the potential of the model's implement-ability under variable temperature and real driving data conditions is verified.

中文翻译：

---

新能源电动汽车锂离子电池动态充电状态捕获的数据驱动新方法

锂离子电池是新能源汽车的主要动力源，准确预测车辆运行过程中未知的荷电状态（SOC）以进行车辆数据监测对于新能源汽车智能化的进步至关重要。在这篇手稿中，提出了一种基于表达树的遗传编程回归模型（ETGPR）来估计锂离子电池的实时SOC。所提出的模型主要采用符号回归技术。除了将电流-电压曲线输入模型之外，还设计了一种额外的方法来确保动态情况下的实时模型预测，其中包括输入参数中前一刻的功率。设置模型中不同的种子超参数，模型自动进行进化计算。随后，对模型的各个参数进行优化调整，经过指定次数的迭代后得到一组准确反映SOC与各个参数之间关系的回归表达式。最后，事实证明，生成的表达式在捕获 SOC 和电池变量之间的非线性关系方面表现更好。此外，与其他模型相比，该模型在存在来自输入无关特征的显着噪声的情况下表现出出色的鲁棒性，均方根误差 (RMSE) 小于 0.3%，平均绝对误差 (MAE) 小于 0.2%实现了。此外，还验证了该模型在变温和真实驾驶数据条件下的可实施性潜力。