The state of charge (SOC) of a battery plays an important role in the battery management system (BMS) of electric vehicles (EVs), since this provides the available runtime for users. However, since driving conditions are various, the monitored battery data (voltage, current, etc.) are also different. If mixed data are used to build an SOC estimation model, the accuracy of the model is low. On the other hand, using only one kind of data set, results in an intelligent model with poor stability and generalization. To resolve these problems, a novel multimode ensemble support vector regression (ME-SVR) method is proposed to estimate SOC. In this method, considering the characters of battery data, the original data set is divided into multiple data subsets by a clustering algorithm. Then, an SVR estimation model is established for each data subset. Finally, the estimation results of multiple SVRs are integrated and the output is obtained according to the weighted average idea of ensemble learning. The experimental results under different driving conditions reveal that this novel algorithm can significantly improve SOC estimation accuracy and enhance the stability and generalization of the model.

电池的荷电状态 (SOC) 在电动汽车 (EV) 的电池管理系统 (BMS) 中起着重要作用，因为它为用户提供了可用的运行时间。但是，由于驾驶条件不同，监测到的电池数据（电压、电流等）也不同。如果使用混合数据建立SOC估算模型，模型的准确度较低。另一方面，仅使用一种数据集会导致智能模型的稳定性和泛化性较差。为了解决这些问题，提出了一种新的多模集成支持向量回归（ME-SVR）方法来估计SOC。该方法考虑到电池数据的特点，通过聚类算法将原始数据集划分为多个数据子集。然后，为每个数据子集建立一个 SVR 估计模型。最后，将多个SVR的估计结果综合起来，根据集成学习的加权平均思想得到输出。不同驾驶条件下的实验结果表明，该算法能够显着提高SOC估计精度，增强模型的稳定性和泛化能力。