为了提高锂离子电池健康状态(state of health,SOH)估计的精确度,本研究结合卷积神经网络(convolutional neural networks,CNN)强大的局部特征提取能力和Transformer的序列处理能力,提出了基于多项式特征扩展的CNN-Transformer融合模型...为了提高锂离子电池健康状态(state of health,SOH)估计的精确度,本研究结合卷积神经网络(convolutional neural networks,CNN)强大的局部特征提取能力和Transformer的序列处理能力,提出了基于多项式特征扩展的CNN-Transformer融合模型。该方法提取了与电池容量高度相关的增量容量(incremental capacity,IC)曲线峰值、IC曲线对应电压、面积及充电时间作为健康因子,然后将其进行多项式扩展,增加融合模型对输入特征的非线性处理能力。引入主成分分析法(principal component analysis,PCA)对特征空间进行降维,有利于捕获数据有效信息,减少模型训练时间。采用美国国家宇航局(National Aeronautics and Space Administration,NASA)数据集和马里兰大学数据集,通过加入多项式特征前后的CNN-Transformer模型对比、加入多项式特征的CNN-Transformer模型和单一模型算法对比,验证了加入多项式特征的CNN-Transformer融合算法的有效性和精确度,结果表明提出模型的SOH估计精度相较于未加入多项式特征的CNN-Transformer模型,对于B0005、B0006、B0007、B0018数据集分别提高了38.71%、50.28%、4.71%、17.58%。展开更多
Damage to elevated water tanks in past earthquakes can be attributed to the poor performance of their supporting frame staging. In order to ascertain the performance of these elevated water tanks, it is crucial to cat...Damage to elevated water tanks in past earthquakes can be attributed to the poor performance of their supporting frame staging. In order to ascertain the performance of these elevated water tanks, it is crucial to categorize the damage in quantifiable damage states. Among various parameters to quantify the damage states, the top drift of frame staging can be conveniently correlated to the different damage levels. In literature, drift limits corresponding to different damage states of the frame staging of the elevated water tank are not available. In the present study, drift limits for RC frame staging in elevated water tanks corresponding to different seismic damage states have been proposed. Various damage states of the elevated water tank have been determined using the Park and Ang damage index. The Park and Ang damage index utilizes results of both pushover analysis and incremental dynamic analysis. Twelve models of elevated water tanks have been developed considering variation in staging height and tank capacity. Incremental dynamic analysis has been performed using the suite of twelve actual earthquake ground motions. Based on the regression analysis between damage indexes and drift, limiting drift values for each damage state are proposed.展开更多
电池健康状态(state of health,SOH)的准确估计是电池管理系统的关键技术之一,对保障电动汽车安全、可靠运行至关重要。针对当前高斯过程回归(gaussian process regression,GPR)中单一核函数泛化性能不足,超参数选取易陷入局部最优导致...电池健康状态(state of health,SOH)的准确估计是电池管理系统的关键技术之一,对保障电动汽车安全、可靠运行至关重要。针对当前高斯过程回归(gaussian process regression,GPR)中单一核函数泛化性能不足,超参数选取易陷入局部最优导致SOH估计精度较低的问题,提出一种灰狼优化算法(grey wolf optimization,GWO)和组合核函数改进GPR的SOH估计方法。首先,基于容量增量分析法提取用于表征电池老化的特征,对电池恒流充电的容量-电压曲线插值并以差分法计算容量增量(increment capacity,IC)曲线,应用Savitzky-Golay滤波平滑处理,提取峰值高度、峰值电压及峰面积作为健康特征;其次,引入多维尺度变换(multidimensional scaling,MDS)消除特征冗余性同时降低模型计算复杂度,利用Pearson系数验证所提健康特征与SOH的相关性;然后,结合SOH退化轨迹的非线性和电池容量再生的准周期性特点,将神经网络核函数与周期核函数组合作为GPR的协方差核函数,以GWO对组合核函数超参数的初值进行优化;最后,基于NASA电池数据集将所提方法与SVR、ELM、GPR模型作对比,检验GWO-GPR模型的准确性,估计结果的最大均方根误差(root mean square error,RMSE)和平均绝对误差(mean absolute error,MAE)分别为1.03%和0.5%,以第60、80、100个循环为估计起始点,验证模型的鲁棒性,结果显示最大RMSE控制在1.03%以内。展开更多
文摘为了提高锂离子电池健康状态(state of health,SOH)估计的精确度,本研究结合卷积神经网络(convolutional neural networks,CNN)强大的局部特征提取能力和Transformer的序列处理能力,提出了基于多项式特征扩展的CNN-Transformer融合模型。该方法提取了与电池容量高度相关的增量容量(incremental capacity,IC)曲线峰值、IC曲线对应电压、面积及充电时间作为健康因子,然后将其进行多项式扩展,增加融合模型对输入特征的非线性处理能力。引入主成分分析法(principal component analysis,PCA)对特征空间进行降维,有利于捕获数据有效信息,减少模型训练时间。采用美国国家宇航局(National Aeronautics and Space Administration,NASA)数据集和马里兰大学数据集,通过加入多项式特征前后的CNN-Transformer模型对比、加入多项式特征的CNN-Transformer模型和单一模型算法对比,验证了加入多项式特征的CNN-Transformer融合算法的有效性和精确度,结果表明提出模型的SOH估计精度相较于未加入多项式特征的CNN-Transformer模型,对于B0005、B0006、B0007、B0018数据集分别提高了38.71%、50.28%、4.71%、17.58%。
文摘Damage to elevated water tanks in past earthquakes can be attributed to the poor performance of their supporting frame staging. In order to ascertain the performance of these elevated water tanks, it is crucial to categorize the damage in quantifiable damage states. Among various parameters to quantify the damage states, the top drift of frame staging can be conveniently correlated to the different damage levels. In literature, drift limits corresponding to different damage states of the frame staging of the elevated water tank are not available. In the present study, drift limits for RC frame staging in elevated water tanks corresponding to different seismic damage states have been proposed. Various damage states of the elevated water tank have been determined using the Park and Ang damage index. The Park and Ang damage index utilizes results of both pushover analysis and incremental dynamic analysis. Twelve models of elevated water tanks have been developed considering variation in staging height and tank capacity. Incremental dynamic analysis has been performed using the suite of twelve actual earthquake ground motions. Based on the regression analysis between damage indexes and drift, limiting drift values for each damage state are proposed.
文摘电池健康状态(state of health,SOH)的准确估计是电池管理系统的关键技术之一,对保障电动汽车安全、可靠运行至关重要。针对当前高斯过程回归(gaussian process regression,GPR)中单一核函数泛化性能不足,超参数选取易陷入局部最优导致SOH估计精度较低的问题,提出一种灰狼优化算法(grey wolf optimization,GWO)和组合核函数改进GPR的SOH估计方法。首先,基于容量增量分析法提取用于表征电池老化的特征,对电池恒流充电的容量-电压曲线插值并以差分法计算容量增量(increment capacity,IC)曲线,应用Savitzky-Golay滤波平滑处理,提取峰值高度、峰值电压及峰面积作为健康特征;其次,引入多维尺度变换(multidimensional scaling,MDS)消除特征冗余性同时降低模型计算复杂度,利用Pearson系数验证所提健康特征与SOH的相关性;然后,结合SOH退化轨迹的非线性和电池容量再生的准周期性特点,将神经网络核函数与周期核函数组合作为GPR的协方差核函数,以GWO对组合核函数超参数的初值进行优化;最后,基于NASA电池数据集将所提方法与SVR、ELM、GPR模型作对比,检验GWO-GPR模型的准确性,估计结果的最大均方根误差(root mean square error,RMSE)和平均绝对误差(mean absolute error,MAE)分别为1.03%和0.5%,以第60、80、100个循环为估计起始点,验证模型的鲁棒性,结果显示最大RMSE控制在1.03%以内。
