基于RVM算法的蓄电池SOC预测  

SOC prediction for battery based on relevance vector machine algorithm

作　　者：侯江娜 孙玉坤[1] 王鹏飞[1] 

机构地区：[1]江苏大学电气信息工程学院,江苏镇江212013

出　　处：《电源技术》2015年第3期523-526,共4页Chinese Journal of Power Sources

摘　　要：由于蓄电池真实的荷电状态与多种因素如电池温度、充放电电压、充放电电流和电池老化等成高度非线性,使得蓄电池荷电状态预测模型建立困难,且预测精度差。针对以上问题,用基于相关向量机的预测方法,以电池充电端电压和充电电流为输入量、电池的荷电状态为输出量建立预测模型,分析该模型性能与高斯核函数带宽之间的关系。通过分析得出,高斯核函数带宽取值为0.9时,相关向量机方法具有较为理想的预测效果。与支持向量机模型相比较,该模型稀疏性强、复杂度低、预测时间短,并提高了对新测试样本点的预测精度,泛化能力强。The actual state of charge for battery was influenced by many factors, such as temperature, charge and discharge voltage, charge and discharge current and aging of battery. It was difficult to build prediction model and the prediction precision was bad. In order to solve these problems, SOC prediction for battery, based on relevance vector, was proposed. Based on charge voltage and charge current as inputs and state of charge（SOC） of battery as output, a relevance vector prediction model was presented, and the relationship between the model performance and the width of Gaussian kernel function was analyzed. According to the analysis, relevance vector machine had a better prediction when width of Gaussian kernel function was 0.9. Compared with the support vector machine（SVM）,results indicate： This model had high sparseness property, reduced the prediction model complexity and prediction time. The prediction precision was enhanced, and high generalization ability was proved.

关 键 词：电池荷电状态 相关向量机 高斯核函数 稀疏性 泛化能力 

分 类 号：TM912[电气工程—电力电子与电力传动]