准确估计锂离子电池(lithium-ion battery,LIB)的健康状态(state of health,SOH)对于确保储能电站的安全稳定运行至关重要.然而,现有的数据驱动方法通常依赖手工特征提取,并且特征的时间尺度比较单一,很难进行高效且精确的电池健康状态评估.为了解决这些问题,提出了一种基于多时间尺度建模自动特征提取和通道注意力机制的健康状态估计模型.该模型首先将充电过程信息输入多个并行的膨胀卷积模块(dilation convolution module,DCM),从不同时间尺度进行自动特征提取,获得丰富且全面的特征表示.随后,不同尺度的特征通过融合后结合门控循环单元(gated recurrent unit,GRU)提取时间序列的长期依赖关系.模型进一步融入通道注意力机制(efficient channel attention,ECA),对历史信息进行相关性动态权重分配,关注显著特征.最后,在两个公开数据集上验证了本方法的优越性,并与其他常用深度学习模型进行了比较.结果表明,本模型具有较高的SOH估计精度和良好的迁移性,两个数据集上的均方根误差分别仅为0.0110和0.0095,在跨数据集的迁移实验中均方误差仅为0.0092.
Estimating lithium-ion battery health using automatic feature extraction and channel attention mechanisms for multi-timescale modeling
Accurate estimation of the state of health(SOH)in lithium-ion batteries(LIB)is crucial for the safe and stable operation of energy storage systems.Current data-driven approaches often rely on manual feature extraction or fall short in single-scale feature representation.To address these issues,this paper introduces a novel SOH estimation model that leverages automatic feature extraction and channel attention mechanisms for multi-timescale modeling.The approach begins with inputting charging process data into multiple parallel dilation convolution modules(DCM),which automatically extract features across various time scales,creating a rich and comprehensive feature representation.These multi-scale features are then integrated and processed by a gated recurrent unit(GRU)to capture long-term dependencies in the time series data.Furthermore,the model incorporates the efficient channel attention(ECA)mechanism,which dynamically adjusts the importance of historical information and emphasizes critical features.The proposed method's effectiveness is validated through experiments two public datasets,showcasing a significant improvement over common deep learning models.Results demonstrate that the model proposed in this study exhibits high precision in SOH estimation and robust transferability.The model achieves low Root Mean Square Errors(RMSE)of 0.0110 and 0.0095 on these datasets,respectively,and maintains an RMSE of only 0.0092 in cross-dataset transfer experiments.These findings underscore the efficacy and adaptability of the proposed model in executing SOH predictions across different datasets.
lithium-ion batterystate of healthconvolutional neural networkattention mechanismtime series
万方数据
维普
