【作者】 朱亚运； 田佳强； 徐瑞龙； 张陈斌；

【Author】 Zhu Yayun;Tian Jiaqiang;Xu Ruilong;Zhang Chenbin;Department of Automation,University of Science and Technology of China;

【机构】 中国科学技术大学；

【摘要】 电池的SOC估计是电池管理系统的最重要的功能之一,是对电动汽车行驶里程的量化评估。由于复杂的电池动态和环境条件,现有的数据驱动电池状态估计技术无法准确估计电池状态。为了克服这个问题,本文通过结合递归神经网络建模和基于粒子滤波的误差消除提出了一种新的SOC估计方法。首先,采用具有长短时间记忆的递归神经网络来学习电池SOC与锂离子电池的可测量变量(例如电流、电压和温度)之间的长期非线性关系。其次,采用粒子滤波对神经网络模型的估计误差进行去噪,来平滑估计结果。本文所提出的方法是无模型的并且能够捕获可测量变量和电池状态之间的长期依赖性。最后,通过在随机工况和不同温度下与传统数据驱动方法比较来验证所提出方法的优越性。更多还原

【Abstract】 The state of charge estimation is the most crucial function of a battery management system,which is the quantified evaluation of driving mileage of electric vehicles.The existing data-driven battery states estimation technologies fail to accurately estimate battery states due to sophisticated battery dynamics and ambient conditions.In order to overcome the issue,this work investigates a data driven-enabled battery states estimation method by a combination of a recurrent neural network modeling and a particle filtering based error cancellation.First,a recurrent neural network with longshort time memory is employed to learn the long-term nonlinear relationship between the battery states and measurable variables of batteries,such as current,voltage,and temperature.Third,in order to reduce the estimation errors of the neural network model,a particle filtering is employed to smooth the estimation results.The proposed method is model-free,which is capable of capture long-term dependence between measurable variables and battery states.Finally,the performance of the proposed method is verified by comparison with conventional data-driven methods under random conditions and at different temperatures.更多还原