摘要
荷电状态SOC(state-of-charge)是电动汽车安全运行和能量管理的关键。传统基于卡尔曼滤波与循环神经网络的SOC估算算法需要一定量的数据来保证估算结果的收敛性,且无法准确估算起始点附近的SOC值。基于此,提出1种使用U-Net结构的卷积神经网络SOC估算方法,该方法可以处理可变长度的输入数据,输出等长的SOC估算结果,同时也能在起始点处准确估算SOC。此外还提出了1种全变分损失函数,在不增加模型复杂度的情况下,仅通过优化损失函数来提高估算的稳定性,并显著降低了最大误差。采用5种恒温条件下的动态驾驶循环数据对该模型进行训练,在恒温和变温条件下的SOC估算结果均具有较高的精度。在恒温条件下,SOC估算的平均绝对误差MAE(mean absolute error)在1.1%以内,均方根误差RMSE(root mean square error)在1.4%以内。在变温条件下,SOC估算的MAE在1.5%以内,RMSE在1.8%以内。
The state-of-charge(SOC)is key to the safe operation and energy management of electric vehicles.The traditional SOC estimation algorithm based on Kalman filter and recurrent neural network requires a period of data to ensure the convergence of estimation results and cannot accurately estimate the SOC value near the starting point.A convolutional neural network SOC estimation method using U-Net structure is proposed,which can deal with variable-length input data and output the equal-length SOC estimation results.Meanwhile,it can also accurately estimate SOC at the starting point.In addition,a full-variance loss function is put forward,which can improve the estimation stability only by optimizing the loss function without increasing the model complexity and significantly reduce the maximum error.This model is trained with dynamic driving cycle data under five constant-temperature conditions,and the SOC estimation results are highly accurate under both the constant-temperature and variable-temperature conditions.Under the constant-temperature condition,the mean absolute error(MAE)and root mean square error(RMSE)of SOC estimation were within 1.1%and 1.4%,respectively.Under the variable-temperature condition,the MAE and RMSE were within 1.5%and 1.8%,respectively.
作者
靳明旭
陈嘉楠
陈宇昊
徐晗
范鑫源
JIN Mingxu;CHEN Jianan;CHEN Yuhao;XU Han;FAN Xinyuan(Locomotive&Car Research Institute,China Academy of Railway Sciences Co.,Ltd.,Beijing 100081,China;State Key Laboratory for Traction and Control System of EMU and Locomotive,Beijing 100081,China;Beijing Zongheng Electro-Mechanical Technology Co.,Ltd.,Beijing 100089,China;School of Electrical Engineering,Beijing Jiaotong University,Beijing 100044,China;CRRC Industrial Research Institute Co.,Ltd.,Beijing 100070,China)
出处
《电源学报》
2025年第7期304-312,共9页
Journal of Power Supply
基金
中国铁道科学研究院集团有限公司科研项目(2020YJ202)。
