基于稀疏采样数据的电动公交车电池SOC预测方法研究

Study on Battery SOC Prediction Method for Electric Bus Based on Sparsely Sampled Data

为提高电动公交车电池SOC预测的精度,基于某电池监控云平台电池数据库中存储的以30 s为采样周期的稀疏采样的电池运行数据,对电动公交车电池SOC预测方法进行了研究。首先,介绍了稀疏采样数据源,分析了电动公交车动力电池的运行过程及其SOC变化的影响因素。选取了当前电池组的总电压、电流、电池模组温度均值及前一时刻SOC值作为预测变量,而选择当前电池组SOC作为输出变量,构建了训练数据集与测试数据集。然后,采用支持向量机(SVM)算法进行训练,并使用贝叶斯优化算法寻找SVM的最优超参数组合,提出了基于稀疏采样数据的电动公交车电池SOC单步预测方法。接着通过对训练数据集的再划分,进一步提出了基于稀疏采样数据的电动公交车SOC自主预测方法,摆脱了在SOC长期预测过程中对于BMS估计的真实SOC值的依赖。试验结果表明,SOC单步预测方法的最大绝对误差仅为1.82%,SOC自主预测方法的最大绝对误差也只有5.89%,都具有较高的预测精度。根据在不同运行路线和不同环境温度下的试验结果,SOC预测模型具有较高的鲁棒性。

In order to enhance the accuracy of SOC prediction for electric bus battery,the SOC prediction method of electric bus battery is studied based on the battery operation data which is sparsely sampled with a sampling period of 30 s and stored on the battery database of a battery monitoring platform.Firstly,the source of sparsely sampled data is introduced,and the operation process of the traction battery of electric bus and the factors affecting the change of SOC are analyzed.The present total voltage,current and the average temperature of battery module and the battery SOC value at previous moment are selected as prediction variables,and the present battery SOC is selected as output variable,so both training and testing data set are constructed.Then support vector machine(SVM)algorithm is adopted for training while Bayesian optimization algorithm is used to find the optimal hyper-parameter combination of SVM,and a single step prediction method for the battery SOC of electric bus based on sparsely sampled data is proposed.Next,by redividing the training data set,an autonomous prediction method for the battery SOC of electric bus based on sparsely sampled data is further put forward for getting rid of the dependence on real SOC value by BMS estimation in the course of long-term SOC prediction.The results of test show that the maximum absolute error with single step SOC prediction is only 1.82%,and that with autonomous SOC prediction is also only 5.89%,both of which have high prediction accuracy.Finally,a test on different routes at different ambient temperatures is conducted with a result indicating that the SOC prediction model adopted has relatively high robustness.