The prediction of excitation band edge wavelength(EBEW)and peak emission wavelength(PEW)for Eu^(2+)-activated phosphors is intricate in practice,although a theoretical interpretation has been well established.A data-d...The prediction of excitation band edge wavelength(EBEW)and peak emission wavelength(PEW)for Eu^(2+)-activated phosphors is intricate in practice,although a theoretical interpretation has been well established.A data-driven approach could be of great help for EBEW and PEW prediction.We collected 91 Eu^(2+)-activated phosphors,the host structures of which exhibit a single activator site and the EBEW and PEW of which are available at the critical activator concentration.We extracted 29 descriptors(input features)that implicate the elemental and structural traits of phosphor hosts,and set up an integrated machine-learning(ML)platform consisting of 18 ML algorithms that allowed prediction of the EBEW and PEW as well as the DFT-calculated band gap(Eg).The acquired dataset involving 91 phosphors was insufficient for the 29-input-feature problem and the real-world data collected from the literature have a so-called dirty nature due to inaccurate,unstandardized experiments.Despite an unavoidable paucity of data and the dirty-data problems of real-world data-based ML implementation,we obtained acceptable holdout dataset test results for PEW predications such as R^(2)>0.6,MSE<0.02,and test_R^(2)/training_R^(2)>0.77 for four ML algorithms.The EBEW and E_(g)predictions returned slightly better test results than these PEW examples.展开更多
为有效提升配电网韧性,提出了一种基于数据-模型混合驱动的多类型移动应急资源优化调度方法。首先,考虑到交通道路状态动态变化对移动储能车(mobile energy storage system,MESS)和应急抢修队(repair crew,RC)策略的影响,构建了以电力-...为有效提升配电网韧性,提出了一种基于数据-模型混合驱动的多类型移动应急资源优化调度方法。首先,考虑到交通道路状态动态变化对移动储能车(mobile energy storage system,MESS)和应急抢修队(repair crew,RC)策略的影响,构建了以电力-交通耦合网总损失成本最小为目标的多类型移动应急资源随机优化调度模型。然后,为了实时准确地求解MESS和RC最优路由和调度策略,提出了一种数据-模型混合驱动方法对所构建的复杂非线性随机优化模型进行求解。在数据驱动部分提出一种图注意力网络多智能体强化学习算法,以求解考虑交通网道路修复时间和移动应急资源邻接关系动态变化等不确定因素的MESS和RC最优路由策略。所提算法有效结合多种改进策略和优先经验回放策略以提高算法的采样效率和训练效果。在模型驱动部分采用二阶锥松弛和大M法将多类型移动应急资源优化调度问题构建为混合整数二阶锥规划模型以求解可再生能源出力和配电网负荷变化影响下MESS和RC最优调度策略。最后,在2个不同规模的电力-交通耦合网中验证所提方法的有效性、泛化能力和可拓展能力。展开更多
锂电池健康状态(state of health,SOH)的在线估计是锂电池管理系统中必不可少的一部分。大部分基于数据驱动的锂电池SOH估计方法由于计算量较大,难以在锂电池管理系统微控制器中在线使用。因此,文中提出基于新型健康特征的锂电池SOH快...锂电池健康状态(state of health,SOH)的在线估计是锂电池管理系统中必不可少的一部分。大部分基于数据驱动的锂电池SOH估计方法由于计算量较大,难以在锂电池管理系统微控制器中在线使用。因此,文中提出基于新型健康特征的锂电池SOH快速估计方法。首先,分析锂电池的充电数据,基于已有的锂电池恒流充电过程的等压升时间(time interval of an equal charging voltage difference,TIECVD)健康特征,构建一个同充电电压起点、同充电时间间隔的健康特征。其次,文中提出基于新型健康特征和多元线性回归(multiple linear regression,MLR)的锂电池SOH快速估计方法。然后,通过对牛津锂电池老化数据集和美国国家航空航天局锂电池随机使用数据集进行分析,以0.01 V步长遍历恒流充电电压区间,以皮尔逊相关系数最大为目标,确定锂电池最优的起始电压。最后,考虑不同充电时间间隔,利用最小二乘(ordinary least squares,OLS)回归分析方法,确定锂电池最优充电时间间隔参数。使用2个数据集划分的训练集建立MLR模型,使用2个数据集划分的验证集对文中方法进行验证。实验结果表明,文中基于新型健康特征方法可极大缩减计算量,并且可以在保障预测精度的前提下实现锂电池SOH的快速估计。展开更多
基金supported by the Creative Materials Discovery Program through the National Research Foundation of Korea(NRF)funded by the Ministry of Science,ICT,and Future Planning(2015M3D1A1069705),(2021R1A2C1011642)and(2021R1A2C1009144)partly by the Alchemist Project(20012196)Digital manufacturing platform(N0002598)funded by MOTIE,Korea.
文摘The prediction of excitation band edge wavelength(EBEW)and peak emission wavelength(PEW)for Eu^(2+)-activated phosphors is intricate in practice,although a theoretical interpretation has been well established.A data-driven approach could be of great help for EBEW and PEW prediction.We collected 91 Eu^(2+)-activated phosphors,the host structures of which exhibit a single activator site and the EBEW and PEW of which are available at the critical activator concentration.We extracted 29 descriptors(input features)that implicate the elemental and structural traits of phosphor hosts,and set up an integrated machine-learning(ML)platform consisting of 18 ML algorithms that allowed prediction of the EBEW and PEW as well as the DFT-calculated band gap(Eg).The acquired dataset involving 91 phosphors was insufficient for the 29-input-feature problem and the real-world data collected from the literature have a so-called dirty nature due to inaccurate,unstandardized experiments.Despite an unavoidable paucity of data and the dirty-data problems of real-world data-based ML implementation,we obtained acceptable holdout dataset test results for PEW predications such as R^(2)>0.6,MSE<0.02,and test_R^(2)/training_R^(2)>0.77 for four ML algorithms.The EBEW and E_(g)predictions returned slightly better test results than these PEW examples.
文摘为有效提升配电网韧性,提出了一种基于数据-模型混合驱动的多类型移动应急资源优化调度方法。首先,考虑到交通道路状态动态变化对移动储能车(mobile energy storage system,MESS)和应急抢修队(repair crew,RC)策略的影响,构建了以电力-交通耦合网总损失成本最小为目标的多类型移动应急资源随机优化调度模型。然后,为了实时准确地求解MESS和RC最优路由和调度策略,提出了一种数据-模型混合驱动方法对所构建的复杂非线性随机优化模型进行求解。在数据驱动部分提出一种图注意力网络多智能体强化学习算法,以求解考虑交通网道路修复时间和移动应急资源邻接关系动态变化等不确定因素的MESS和RC最优路由策略。所提算法有效结合多种改进策略和优先经验回放策略以提高算法的采样效率和训练效果。在模型驱动部分采用二阶锥松弛和大M法将多类型移动应急资源优化调度问题构建为混合整数二阶锥规划模型以求解可再生能源出力和配电网负荷变化影响下MESS和RC最优调度策略。最后,在2个不同规模的电力-交通耦合网中验证所提方法的有效性、泛化能力和可拓展能力。
文摘锂电池健康状态(state of health,SOH)的在线估计是锂电池管理系统中必不可少的一部分。大部分基于数据驱动的锂电池SOH估计方法由于计算量较大,难以在锂电池管理系统微控制器中在线使用。因此,文中提出基于新型健康特征的锂电池SOH快速估计方法。首先,分析锂电池的充电数据,基于已有的锂电池恒流充电过程的等压升时间(time interval of an equal charging voltage difference,TIECVD)健康特征,构建一个同充电电压起点、同充电时间间隔的健康特征。其次,文中提出基于新型健康特征和多元线性回归(multiple linear regression,MLR)的锂电池SOH快速估计方法。然后,通过对牛津锂电池老化数据集和美国国家航空航天局锂电池随机使用数据集进行分析,以0.01 V步长遍历恒流充电电压区间,以皮尔逊相关系数最大为目标,确定锂电池最优的起始电压。最后,考虑不同充电时间间隔,利用最小二乘(ordinary least squares,OLS)回归分析方法,确定锂电池最优充电时间间隔参数。使用2个数据集划分的训练集建立MLR模型,使用2个数据集划分的验证集对文中方法进行验证。实验结果表明,文中基于新型健康特征方法可极大缩减计算量,并且可以在保障预测精度的前提下实现锂电池SOH的快速估计。
