基于LSTM神经网络的动力电池SOC估算研究  

SOC ESTIMATION OF POWER BATTERY BASED ON LSTM NEURAL NETWORK

作　　者：郑永飞 文怀兴[1] 韩昉[1] 杨鑫 张晶[3] Zheng Yongfei;Wen Huaixing;Han Fang;Yang Xin;Zhang Jing(College of Mechanical and Electrical Engineering,Shaanxi University of Science and Technology,Xi’an 710021,Shaanxi,China;Xi’an GuanTong Digital Source Electronics Co.,Ltd,Xi’an 710065,Shaanxi,China;Xi’an Institute of Applied Optics,Xi’an 710065,Shaanxi,China)

机构地区：[1]陕西科技大学机电工程学院,陕西西安710021 [2]西安冠通数源电子有限公司,陕西西安710065 [3]西安应用光学研究所,陕西西安710065

出　　处：《计算机应用与软件》2020年第3期78-81,88,共5页Computer Applications and Software

基　　金：咸阳市二○一八年科学技术研究计划项目(2018k02-16)。

摘　　要：电池监控是电动汽车安全行驶的必要手段,电池的荷电状态(State of Charge,SOC)则是衡量电动汽车安全性能的直接指标。针对锂离子电池的非线性特性,设计一种基于深度学习的长短期记忆网络(Long Short-Term Memory,LSTM)SOC预测模型。通过MATLAB实验验证以及与其他算法的比较分析得出,该模型可以有效解决传统神经网络容易陷入局部最小值以及出现梯度消失、爆炸等问题,估算误差小于2%,具有较高的精度和应用前景。Battery monitoring is a necessary means for safe driving of electric vehicles.State of Charge(SOC)of batteries is a direct indicator to measure the safety performance of electric vehicles.In view of the nonlinear characteristics of lithium-ion batteries,this paper designs a long short-term memory(LSTM)SOC prediction model based on deep learning.Through the experimental verification of MATLAB and comparative analysis with other algorithms,it is concluded that this model can effectively solve the problems of traditional neural network,such as easily falling into local minimum value,gradient disappearance,explosion and so on.The estimation error is less than 2%,and it has high accuracy and application prospect.

关 键 词：动力电池 LSTM 荷电状态 MATLAB 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]