LiFePO_(4),one of the mainstream cathode materials of current EV batteries,exhibits experimental diffusion coefficients(D_(c))of Li^(+)which are not only several orders of magnitude lower than those predicted by the i...LiFePO_(4),one of the mainstream cathode materials of current EV batteries,exhibits experimental diffusion coefficients(D_(c))of Li^(+)which are not only several orders of magnitude lower than those predicted by the ionic hopping barriers obtained from theoretical calculations and spectroscopic measurements,but also span several orders from 10^(-14)to 10^(-18)cm^(2)s^(-1)under different states of charge(SOC)and the charging rates(C-rates).Atomic level understanding of such sluggishness and diversity of Li^(+)transport kinetics would be of significance in improving the rate performance of LiFePO_(4)through material and operation optimization but remain challenging.Herein,we show that the high sensitivity of Li^(+)hopping barriers on the local Li–Li coordination environments(numbers and configurations)plays a key role in the ion transport kinetics.This is due a neural network-based deep potential(DP)which allows accurate and efficient calculation of hopping barriers of Li^(+)in LiFePO_(4)with various Li–Li coordination environments,with which the kinetic Monte-Carlo(KMC)method was employed to determine the D_(c)values at various C-rates and SOC across a broad spectrum.Especially,an accelerated KMC simulation strategy is proposed to obtain the D_(c)values under a wide range of SOC at low C-rates,which agree well with that obtained from the galvanostatic intermittent titration technique(GITT).The present study provides accurate descriptions of Li^(+)transport kinetics at both very high and low C-rates,which remains challenging to experiments and first-principles calculations,respectively.Finally,it is revealed that the gradient distributions of Li^(+)density along the diffusion path result in great asymmetry in the barriers of the forward and backward hopping,causing very slow diffusion of Li^(+)and the diverse variation of D_(c).展开更多
Can physical concepts and laws emerge in a neural network as it learns to predict the observation data of physical systems? As a benchmark and a proof-of-principle study of this possibility, here we show an introspect...Can physical concepts and laws emerge in a neural network as it learns to predict the observation data of physical systems? As a benchmark and a proof-of-principle study of this possibility, here we show an introspective learning architecture that can automatically develop the concept of the quantum wave function and discover the Schr?dinger equation from simulated experimental data of the potential-todensity mappings of a quantum particle. This introspective learning architecture contains a machine translator to perform the potential to density mapping, and a knowledge distiller auto-encoder to extract the essential information and its update law from the hidden states of the translator, which turns out to be the quantum wave function and the Schr?dinger equation. We envision that our introspective learning architecture can enable machine learning to discover new physics in the future.展开更多
本文基于美的大数据云平台中采集的多联机系统运行数据,分析了成都市某商业办公建筑多联机系统的运行现状,发现其存在较多设定温度过低的现象,导致能耗过高。为此,提出了将设定温度提高至26℃的节能策略。为了准确计算提高设定温度后的...本文基于美的大数据云平台中采集的多联机系统运行数据,分析了成都市某商业办公建筑多联机系统的运行现状,发现其存在较多设定温度过低的现象,导致能耗过高。为此,提出了将设定温度提高至26℃的节能策略。为了准确计算提高设定温度后的节能潜力,采用4种机器学习算法建立了室温预测模型。对比算法发现支持向量回归(SVR)模型的精度最高,其3台内机的室内温度平均绝对误差MAE误差仅为0.094℃、0.189℃和0.127℃。最后,基于模型预测的室内温度,结合机组的能耗特性曲线,发现将设定温度提高到26℃后,系统的整体能耗从2551.99 kW·h下降到1579.64 k W·h,节能率达到了38.1%。展开更多
基金financially supported by the National Natural Science Foundation of China(22272122,21832004 and 21673163)。
文摘LiFePO_(4),one of the mainstream cathode materials of current EV batteries,exhibits experimental diffusion coefficients(D_(c))of Li^(+)which are not only several orders of magnitude lower than those predicted by the ionic hopping barriers obtained from theoretical calculations and spectroscopic measurements,but also span several orders from 10^(-14)to 10^(-18)cm^(2)s^(-1)under different states of charge(SOC)and the charging rates(C-rates).Atomic level understanding of such sluggishness and diversity of Li^(+)transport kinetics would be of significance in improving the rate performance of LiFePO_(4)through material and operation optimization but remain challenging.Herein,we show that the high sensitivity of Li^(+)hopping barriers on the local Li–Li coordination environments(numbers and configurations)plays a key role in the ion transport kinetics.This is due a neural network-based deep potential(DP)which allows accurate and efficient calculation of hopping barriers of Li^(+)in LiFePO_(4)with various Li–Li coordination environments,with which the kinetic Monte-Carlo(KMC)method was employed to determine the D_(c)values at various C-rates and SOC across a broad spectrum.Especially,an accelerated KMC simulation strategy is proposed to obtain the D_(c)values under a wide range of SOC at low C-rates,which agree well with that obtained from the galvanostatic intermittent titration technique(GITT).The present study provides accurate descriptions of Li^(+)transport kinetics at both very high and low C-rates,which remains challenging to experiments and first-principles calculations,respectively.Finally,it is revealed that the gradient distributions of Li^(+)density along the diffusion path result in great asymmetry in the barriers of the forward and backward hopping,causing very slow diffusion of Li^(+)and the diverse variation of D_(c).
基金financially supported by the National Key Research and Development Program of China (2016YFA0301600)the National Natural Science Foundation of China (11734010)the support of the China Scholarship Council
文摘Can physical concepts and laws emerge in a neural network as it learns to predict the observation data of physical systems? As a benchmark and a proof-of-principle study of this possibility, here we show an introspective learning architecture that can automatically develop the concept of the quantum wave function and discover the Schr?dinger equation from simulated experimental data of the potential-todensity mappings of a quantum particle. This introspective learning architecture contains a machine translator to perform the potential to density mapping, and a knowledge distiller auto-encoder to extract the essential information and its update law from the hidden states of the translator, which turns out to be the quantum wave function and the Schr?dinger equation. We envision that our introspective learning architecture can enable machine learning to discover new physics in the future.
基金financially supported by the National Key R&D Program of China (2022YFB3807200)Shanghai Explorer Program (Batch I) (23TS1401500)+2 种基金the National Natural Science Foundation of China (22133005)the Project funded by China Postdoctoral Science Foundation (2022M723276 and GZB20230793)Shanghai Sailing Program (23YF1454900), and Shanghai Post-doctoral Excellence Program (2022660)。
文摘本文基于美的大数据云平台中采集的多联机系统运行数据,分析了成都市某商业办公建筑多联机系统的运行现状,发现其存在较多设定温度过低的现象,导致能耗过高。为此,提出了将设定温度提高至26℃的节能策略。为了准确计算提高设定温度后的节能潜力,采用4种机器学习算法建立了室温预测模型。对比算法发现支持向量回归(SVR)模型的精度最高,其3台内机的室内温度平均绝对误差MAE误差仅为0.094℃、0.189℃和0.127℃。最后,基于模型预测的室内温度,结合机组的能耗特性曲线,发现将设定温度提高到26℃后,系统的整体能耗从2551.99 kW·h下降到1579.64 k W·h,节能率达到了38.1%。
