关键词: 电池荷电状态, 稀疏采样数据, 支持向量机, 贝叶斯优化算法

Abstract: 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.

Key words: SOC, sparsely sampled data, support vector machine, Bayesian optimization algorithm