基于BI-LSTM神经网络的宽采样频率电池SOH估算

Battery SOH Estimation with Wide Sampling Frequency Based on BI-LSTM Neural Network

锂离子电池健康状态(SOH)直接决定了电池储存能量和输出功率的能力。搭载锂离子电池的交通工具在运行时,需要实时上传电池数据,数据记录频率越高,数据通信成本越高。为了保证电池SOH估算准确,同时降低数据通信成本,基于试验室环境,设计了不同充放电倍率下的宽采样频率充放电试验。为了解决宽采样频率下健康特征波动问题,采用局部加权线性回归(LWLR)算法对健康特征下降趋势定性刻画。采用最大信息系数(MIC)算法衡量健康特征与容量的相关性。最后,基于双向长短期记忆(BI-LSTM)神经网络进一步学习容量与健康特征的非线性退化关系。根据单节电池历史数据离线估算电池SOH,最大相对误差为1.601%。

The state of health(SOH) of lithium-ion battery directly determines the ability to store energy and output power.When the transportation equipped with lithium-ion battery is running,the battery data needs to be uploaded in real time.The higher the data recording frequency,the higher the data communication cost.In order to ensure the accuracy of battery SOH estimation and reduce the data communication cost,a wide sampling frequency charge-discharge experiment with different charge-discharge rates was designed.For the fluctuation of health features of wide sampling frequency,locally weighted linear regression(LWLR) algorithm was used to qualitatively characterize the health features decline trend.Maximum information coefficient(MIC) algorithm was used to measure the correlation between health features and capacity.Finally Bi-directional long and short-term memory(BI-LSTM) based neural network further learned the nonlinear degradation relationship between capacity and health features.Estimating the battery SOH offline was conducted based on the single battery historical data,and the maximum relative error was 1.601%.