摘要
准确估计电池健康状态是设备稳定运行的关键。针对当前健康状态研究中容量难以直接测量、估计模型调参费时等问题,提出基于多健康特征的贝叶斯优化(BO)算法优化卷积神经网络(CNN)与双向长短期记忆(BiLSTM)神经网络预测模型。基于NASA公开锂电池数据,提取3种健康特征。将CNN与BiLSTM结合,提高时间序列数据处理能力,加入BO算法自动搜寻最优参数集,避免组合网络模型陷入局部最优,从而减少评估时间。对比分析相关神经网络模型,结果表明所提方法预测准确度最高,可有效估计锂电池的健康状态,平均绝对误差和方均根误差均在1%以内。
Accurate estimation of battery state of health(SOH)is the key to the stable operation of the device.In order to solve the problems in the current SOH research,such as the difficulty to measure the volume directly and the time required to adjust the model parameters,a prediction model based on the multi-health features of Bayesian optimization(BO)optimized convolution neural network(CNN)and bi-directional long short term memory(BiLSTM)neural network is proposed.Based on NASA’s publicly available lithium battery data,three health characteristics are extracted.The combination of CNN and BiLSTM improves the processing ability of time series data,and adds BO algorithm to automatically search the optimal parameter set,which avoids the combination network model falling into the local optimal and reduces the estimation time.The results show that the proposed method has the highest prediction accuracy and can aeffectively estimate the SOH of lithium batteries.The mean absolute error and root mean square error are both within 1%.
作者
衣思彤
刘雅浓
马耀浥
李文婕
孔航
YI Sitong;LIU Yanong;MA Yaoyi;LI Wenjie;KONG Hang(School of Automation and Electrical Engineering,Dalian Jiaotong University,Dalian,Liaoning 116028;School of Rolling Stock Engineering,Dalian Jiaotong University,Dalian,Liaoning 116028;School of Computer and Communication Engineering,Dalian Jiaotong University,Dalian,Liaoning 116028)
出处
《电气技术》
2024年第5期1-10,21,共11页
Electrical Engineering
基金
辽宁省自然科学基金(2021-MS-298)。
