融合MHSA的锂电池SOC估计  

FUSION OF MHSA LITHIUM BATTERY SOC ESTIMATES

作　　者：高俊东 马志强 刘广忱[2,3] 宝财吉拉呼 李宏勋[1] 刘宇龙 Gao Jundong;Ma Zhiqiang;Liu Guangchen;Bao Caijilahu;Li Hongxun;Liu Yulong(College of Data Science and Application,Inner Mongolia University of Technology,Hohhot 010080,China;Engineering Research Center of Large-scale Energy Storage Technology,Ministry of Education,Hohhot 010080,China;College of Electric Power,Inner Mongolia University of Technology,Hohhot 010080,China)

机构地区：[1]内蒙古工业大学数据科学与应用学院,呼和浩特010080 [2]大规模储能技术教育部工程研究中心,呼和浩特010080 [3]内蒙古工业大学电力学院,呼和浩特010080

出　　处：《太阳能学报》2025年第3期16-24,共9页Acta Energiae Solaris Sinica

基　　金：国家自然科学基金(51867020);内蒙古自治区高等学校碳达峰碳中和研究项目(STZX202307);内蒙古自治区研究生科研创新项目(KC2024044S)。

摘　　要：提出在长短期记忆网络(LSTM)基础上融合多头自注意力机制(LSTM-MHSA)的方法估计动力电池荷电状态(SOC),将动力电池的电流、电压和温度作为LSTM模型输入,通过带有记忆功能的门控单元选择和遗忘历史数据特征;LSTM处理长序列数据会将所有数据特征同等对待,不能对SOC影响较大的数据特征有效表示,MHSA单元根据数据特性确定特征关注的焦点,提升模型在时序数据中全局和局部的特征提取能力,改进LSTM在对历史数据特征关注度不足和提高模型估计精度;最后在公开数据集上通过对比和泛化实验。结果显示该方法在平均绝对误差上减少0.007,验证在复杂非线性关系的影响下该方法能有效表示长序列数据特征,进而提高SOC估计的精度。A method based on long short-term memory network(LSTM)and multi-head Self-attention mechanism(LSTM-MHSA)was proposed to estimate state of charge(SOC)of power battery.The current,voltage and temperature of the power battery are used as inputs to the LSTM model,and the historical data features are selected and forgotten through a gating unit with memory function.LSTM processes all data features equally in terms of long sequence data and cannot effectively represent data features with greater impacts on SOC.MHSA unit is used to determine the feature focus according to the data characteristics,aiming at enhancing the global and local feature extraction ability of the model in the time series data and improving the LSTM’s insufficient focus on historical data features,therefore improve the model′s estimation accuracy.Finally,the comparison and generalization experiments on public datasets indicate that the method reduces the average absolute error by 0.007,which proves that the method can effectively demonstrate long series data features under the influence of complex nonlinear relationships,thus improve the accuracy of SOC estimation.

关 键 词：锂电池 电动汽车 电池储能 长短期记忆网络 荷电状态 

分 类 号：TM912[电气工程—电力电子与电力传动] U469.72[机械工程—车辆工程]