无机材料学报, 2022, 37 (12): 1321, 网络出版: 2023-01-12  

基于机器学习的BiFeO3-PbTiO3-BaTiO3固溶体居里温度预测

Curie Temperature Prediction of BiFeO3-PbTiO3-BaTiO3 Solid Solution Based on Machine Learning
作者单位
摘要
钙钛矿(ABO3)型压电陶瓷的发展已有几十年历史, 现存有大量数据, 从这些数据中寻找出材料结构与性能之间的关系很有意义。本工作收集了BiFeO3-PbTiO3-BaTiO3钙钛矿型压电陶瓷居里温度(Tc)实验数据, 通过机器学习,构建钙钛矿型压电陶瓷Tc的预测模型。热力学角度, Tc与约合质量符合二次多项式关系, 但偏差较大。选择元素信息、物理量、空间群编号等基础描述符, 利用基于压缩感知原理的SISSO(Sure Independence Screening and Sparsifying Operator)方法进行机器学习, 找出了Tc与成分之间的相关性。比较不同描述符在不同维度上的均方根误差RMSE (Root Mean Square Error), 发现描述符越多、越基础, 维数越大、RMSE越小。同时比较相同个数描述符在同一维度下的RMSE, 用约合质量、A位和B位的离子半径比、A位和B位的未填充电子数比和Ba、Pb、Bi的元素含量等六个描述符构建出最优的四维模型, 其RMSE为0.59 ℃, 最大绝对误差(MaxAE)为1.38 ℃, 外部测试的平均相对误差MRE (Mean Relative Error)为1.00%。结果表明,利用SISSO可以进行有限样本钙钛矿型压电陶瓷Tc的机器学习预测。
Abstract
Perovskite (ABO3) piezoceramics have been developed for several decades, and there are a lot of data available. It is of great significance to find relationships between structure and properties of materials from these data. In this work, experimental data of Curie temperature (Tc) of BiFeO3-PbTiO3-BaTiO3 solid solution of perovskite piezoelectric ceramics was collected to build the model to predict the Tc. From the perspective of thermodynamics, the quadratic polynomial relationship between Tc and reduced mass was introduced but the deviation was relatively large. More descriptors (including element information, physical quantities, space groups number) and SISSO (Sure Independence Screening and Sparsifying Operator) were used for machine learning to find the correlation between Tc and components. Comparing the root mean square error (RMSE) of different descriptors and dimensions, it's found that more descriptors, more fundamental the descriptors are, and larger dimension will result in smaller RMSE to be used. Meanwhile, RMSE of the same number of descriptors in the same dimension are compared. The optimal four-dimensional model is build using six descriptors: reduced mass, the ratio of A- and B-site ion radii, the ratio of A- and B-site unfilled electrons and element contents of Ba, Pb and Bi. RMSE and maximum absolute error (MaxAE) of our model are 0.59 ℃ and 1.38 ℃, respectively. The average relative error (MRE) of external test is 1.00%. Our results indicate that SISSO machine learning based on limited samples is suitable for the predication of Tc of perovskite piezoelectric ceramics.

钙钛矿结构BiFeO3-PbTiO3(BF-PT)基压电陶瓷具有准同型相界(MPB)与高居里温度(Tc), 在智能机电系统传感器、滤波器和驱动器等方面有广泛的应用前景[1-4]图1是典型ABO3钙钛矿结构示意图[5], 通常, A位是低价、半径较大的离子; B位为高价、半径较小的离子, 处于氧八面体中心。居里温度Tc是ABO3 钙钛矿铁电/压电陶瓷的重要参数, 它表示铁电相与顺电相之间转变温度。压电陶瓷的制备通常采用固相反应法, 用材多、耗时长。因此,从实验上获得的Tc数据集较小较稀疏。传统研究在寻找合适Tc压电陶瓷的过程中, 需要反复进行实验验证。这种实验室试错的方法不适用研究组分较多的压电陶瓷体系, 因为每增加一种新的元素, 样本空间将呈现数量级式地增加。

图 1. ABO3钙钛矿结构示意图

Fig. 1. Schematic diagram of ABO3 perovskite structure

下载图片 查看所有图片

近年来, 数据驱动的机器学习(Machine Learning, ML)材料研究方法引起了学界的广泛关注[6-9]。ML越来越多地被应用于分子科学、材料科学和生物科学等领域[10-12]。数据驱动+机器学习的研究范式显著加速了探索复杂陶瓷样本空间的材料研究。与传统实验或演绎方法不同, ML通过正交展开、模式识别、人工神经网络和遗传算法等方式对材料进行优化设计并取得了良好的效果[13-16]。然而, 关于钙钛矿结构压电材料的宏观极化与组分关系方面的仿真设计结果仍差强人意[17]; 利用高通量密度泛函理论筛选高性能压电材料在准确性等方面也值得商榷[18]

Lookman等[19]小数据集主动学习模型加速了目标性能陶瓷的实验合成, 具体步骤包括: (i)训练已有数据集并建立预测模型; (ii)预测最高性能未知体系的组分; (iii)实验验证该组分的性能; (iv)把新的数据加入原始数据集, 并重新训练得到新的模型; (v)重复以上过程若干次, 直到获得足够满意的高性能陶瓷。这套方法适用于目标性能陶瓷的快速设计。然而, 一个非常重要的技术问题是构建小数据集机器学习模型时, 如何选择好的特征描述符?Askanazi等[20]对比了一系列“黑箱”回归算法(如K-近邻算法、支持向量机、随机森林等), 并指出输入回归模型的特征变量的选择对小数据集模型非常重要。

OUYANG等[21] 基于压缩感知理论提出了一套SISSO (Sure Independence Screening and Sparsifying Operator)特征工程研究框架[22-23]。该方法从基础物理量出发, 构造大量稀疏的高维描述符, 然后根据关联性降维筛选最有效的描述符, 并给出描述符与目标之间的函数关系。他们应用SISSO得到了不同氧化态和任意配位时的离子半径, 扩展了Shannon离子半径数据库。

本研究以课题组BiFeO3-PbTiO3-BaTiO3 (BF- PT-BT)三元系固溶体压电陶瓷的Tc实验数据为对象, 利用SISSO进行数据挖掘, 建立材料组分和Tc之间的量化关系。首先, 从热力学角度, 采用约合质量作为反映组分变化的描述符; 然后, 根据化学式构造更多基础描述符, 通过机器学习筛选描述符并得出合适的预测模型; 最后, 通过外部验证预测模型的可靠性。最终建立的一种多维度、多描述符、反映组分和性能关系的量化模型能够快速、简单、有效地预测钙钛矿结构压电陶瓷的居里温度。

1 数据集和描述符

首先, 选取22个BF-PT-BT三元系压电陶瓷的组分和居里温度实验数据[24], 建立内部训练集; 再选取3个相同体系实验数据作为外部测试集[25], 这些数据均在同一实验环境下采用传统的固相反应法制得。通过测量材料的介电常数随温度升高的变化, 将100 kHz频率下的介电常数变化的峰值对应的温度作为居里温度。居里温度为材料的本征参数与组分的关联性较大, 因此机器学习预测的准确率会较高且实用性较强。

Goldschmidt容忍因子作为特征参数被应用在描述钙钛矿晶体和陶瓷材料上已约100年, 现被大量回归算法机器学习模型所使用。然而该容忍因子主要用于描述氧转模式和铁电模之间的关联性, 或者说结构相关性质, 其本身就是基础特征rArB(A位和B位离子半径)的高维复合的一种数学表示。但该容忍因子在描述居里温度上存在一定的局限性。基于SISSO方法可以自动地构造出比它更优异的“新容忍因子”, 在判断是否为钙钛矿结构的精度可以达到92%, 而Goldschmidt容忍因子只有近75%。针对BF-PT-BT固溶体居里温度预测, 本工作只选择了rArB作为基础特征, 而构造类似容忍因子的过程则交给SISSO自动实现。最终选出了4个最重要的高维变量(即xi的函数)并用其构造出了回归模型函数y, 后续章节将验证该高维变量模型的精度。

表1给出选择的特征量及其物理意义, 包括元素信息、物理量、空间群编号在内共12个基础描述符[27]。其中, μ为原胞约合质量, AB分别为ABO3 钙钛矿中占据A、B位元素性质的组分加权平均值。除μ和元素含量外, 其余特征量均采用A和B比值的形式, 以体现出A位和B位元素的相对性质。

表 1. 基础描述符及其物理意义

Table 1. Basic descriptor and related physical meaning

FeatureNamePhysical attributes
μReduced massReduced mass of atoms
A/BIonic radiusShannon ion radius
A/B_CCovalent radiusThe covalent bond radius of an atom
A/B_EElectronegativityElectronegativity of atoms
A/B_NVNValenceThe number of electrons of unfilled orbitals
A/B_NUNUfilledThe number of electrons of unfilled orbitals
A/B_SSpace group numberingThe serial number of the element's space group in the space group table
Ti,Fe,Ba,Pb,BiElement contentElement content ratio of metal ions

查看所有表

2 数据处理和机器学习

从热力学角度, 钙钛矿结构氧化物原胞的涨落可以用约合质量μ来描述[28], 用下列公式计算:

μ1=i=A,B,3O1mimi=(A or B)=(1jxj)mM0+jxjmMj

其中, mMj(j=0, 1,…, n)代表占据A或B位第j个元素的相对原子质量, xj为对应元素含量。将钙钛矿氧化物的组成变化归纳为一个参量μ, 可以简化组分与性能之间关系。本研究首先对BF-PT-BT三元固溶体居里温度实验数据Tc与约合质量μ以及化学压μ*A/B关系进行拟合, 获得了组分和Tc的量化关系; 然后引入更多特征量, 采用SISSO筛选出最合适的基础描述符并确定描述符和模型的维度。最后, 通过外部数据集验证模型的可靠性。

3 结果和讨论

3.1 二次多项式拟合

图2给出了Tcμμ*A/B的关系。彩色点表示实验数据, 红色实线是根据文献报道的二次多项式的拟合结果[28]。可以看出, 实验结果和拟合结果总体是一致的, 即随μμ*A/B增大, Tc分别上升和下降, 具体到组分, 即BT和PT含量增加导致Tc下降。以μ为自变量, 二次多项式拟合结果的决定系数R2更大, 拟合效果更好。μ*A/B虽然提升了描述符的复杂程度, 但是拟合结果误差反而增大。总的说来, 仅用μμ*A/B作为描述符的拟合结果误差是比较大的。

图 2. Tcμμ*A/B的关系

Fig. 2. Relationship between Tc and μ or μ*A/B

下载图片 查看所有图片

3.2 特征变量选择

为降低预测误差, 引入表1中更多的基础描述符。筛选掉冗余的描述符, 帮助降低输入空间维数, 同时尽可能地保证了信息的完整性[29]图3显示了Tc和候选描述符之间的相关性(Correlation, 用皮尔森相关系数R表示)[30]。可以看出, μA/B_E、A/B_NV、A/B_NU、Fe、Pb、Bi和Tc呈正相关, A/BA/B_C、A/B_S、Ti、Ba则和Tc呈负相关; μA/B是与Tc相关性最大的两个描述符。结合候选描述符与Tc之间的相关性系数, 进一步利用SISSO得到的RMSE[31]来选择最终的基础描述符。RMSE越小, 表示拟合程度越好。输出公式中描述符出现的次数可用来判断特征量的重要性。R和RMSE定义如下:

R=ni=1nxieii=1nxii=1neini=1nxi2i=1nxi2ni=1nei2i=1nei2RMSE=1ni=1n(piei)

其中, xi是第i个样本的描述符参数, piei分别为Tc的预测值和实验值, n是样本数量。

图 3. Tc与候选描述符的相关性

Fig. 3. Correlations between the Tc and the candidate descriptors

下载图片 查看所有图片

图4给出了不同运算方式下, 以μμ*A/B为描述符训练模型的RMSE和MaxAE随维度变化的趋势。运算方式1代表加、减、乘、除、倒数、平方和立方; 运算方式2在运算方式1基础上增加了绝对值、自然指数、对数、平方根和立方根。SISSO预测模型一般不超过四维, 即多项式项数加上常数项不超过五项。维数越多, 一方面会过拟合, 另一方面会占用更多的运算时间和资源。比较图4(a~d), 发现维度越大, 运算方式越复杂, RMSE和MaxAE越低, 这说明高维度和复杂运算方式能产生更好的组合描述符, 充分挖掘和利用描述符的信息。从图4还可以看出, 仅用一个描述符时, 描述符越复杂, RMSE越大, 说明蕴含过多信息的描述符会对运算产生较大干扰。总之, 选择更基础的描述符和运用更复杂的运算方式得到的拟合结果误差更小。

图 4. 不同运算方式下μμ*A/B随维度变化的RMSE和MaxAE

Fig. 4. RMSE and MaxAE of μ and μ*A/B varied with dimension under different operation modes

下载图片 查看所有图片

采用运算方式2, 图5给出了两个描述符(图5(a))和三个描述符(图5(b))的拟合误差随维度的变化趋势。对比图5图4(a, c), 发现使用多个基础描述符比组合描述符的输出误差更低。

图 5. 采用两个描述符(a)和三个描述符(b)拟合结果的RMSE和MaxAE随维度的变化

Fig. 5. RMSE and MaxAE of two descriptors (a) and three descriptors (b) as a function of dimension

下载图片 查看所有图片

图6反映了描述符的变化和模型维度对拟合结果RMSE和MaxAE的影响。图6(a, b)分别是以μA/B为前两个描述符改变第三个描述符, 和以μA/B、Ba、Pb和Bi为前五个描述符改变第六个描述符, 对RMSE和MaxAE的影响。添加更多描述符明显降低了预测模型误差。第三和第六个描述符的变化对RMSE的影响较小, 而对MaxAE的影响较大。这说明不同描述符在运算过程中有交互作用, 在共同减小整体误差的同时, 对局部误差影响可能增大。值得注意的是, A/B_NU分别作为第三和第六个描述符时, 拟合结果的RMSE和MaxAE都是最小的。因此, 除了μA/B外, A/B_NU对材料Tc的影响较大, 这与图3A/B_NU描述符相关性系数较大是一致的。图6(c)表明, 从初始两个描述符逐步添加至7个描述符, RMSE逐渐降低, 添加第三个描述符, 在描述符的交互作用下, MaxAE有所提高。总之, 选择多个相对独立的基础特征量为描述符有助于降低拟合结果的误差。7个描述符时, 运算时间延长了1个数量级, 但拟合误差没有明显改善。最终, 选择六个描述符来拟合模型是最优的。图6(d) 显示了最终选择的六个基础描述符模型的RMSE和MaxAE随维度的变化。

图 6. 描述符的变化和模型维度对拟合结果RMSE和MaxAE的影响

Fig. 6. Effects of the changes of descriptor and model dimension on the RMSE and MaxAE

下载图片 查看所有图片

3.3 模型的建立与验证

表2给出了正则化的外部测试集数据, 其中, μʹ和A/Bʹ分别为(μ-4.7)×10和(A/B-2.2)×10。拟合得出的预测模型如下:

y=a(x4x1)/(x3x4)x13(x5+x1)+b(x3x1)(x2x3)(x2x6)(x3x5)+cx1x43x6x5x2x1+dx22/(x2x5)(x3x5)x1x2+e

其中, a=1.977447895, b=2.964590758, c= 1.778816427, d=0.063342758, e=647.6695156, xi是基础描述符。该预测模型的RMSE为0.59 ℃, MaxAE为1.38 ℃。

表 2. 调整描述符参数后的测试集

Table 2. The test sets after adjusting descriptor parameters

SampleTc/℃ (y) μʹ (x1) A/ (x2) A/B_NU (x3) Ba (x4) Pb (x5) Bi (x6)
Bi0.62Pb0.23Ba0.15Fe0.62Ti0.38O35470.28210.38820.50360.07500.11500.3100
Bi0.6Pb0.25Ba0.15Fe0.6Ti0.4O35400.26900.46060.50000.07500.12500.3000
Bi0.54Pb0.31Ba0.15Fe0.54Ti0.46O35020.22950.67890.48970.07500.15500.2700

查看所有表

模型中出现次数最多的为x1x2x3(μA/BA/B_NU), 分别出现了7、6和5次, 说明μA/BA/B_NU的确对Tc影响较大, 这与相关性和SISSO的误差分析结果一致。图7为训练集和测试集实验值和预测Tc值。表3给出测试集的外部验证结果, 平均相对误差(MRE)为1.00%。测试集与总数据集的预测Tc与实验Tc的相关系数R分别为0.99998和0.99995。其中MRE的定义如下:

MRE=1ni=1npieiei×100%

表 3. 外部验证集结果

Table 3. Results of external verification set

SampleExperim-ental Tc/℃ Predi-ction Tc/℃ Abso-lute error/℃ Rela-tive-error/%
Bi0.62Pb0.23Ba0.15Fe0.62Ti0.38O3547544.912.090.38
Bi0.6Pb0.25Ba0.15Fe0.6Ti0.4O3540536.123.880.72
Bi0.54Pb0.31Ba0.15Fe0.54Ti0.46O3502492.529.481.89

查看所有表

图 7. 训练集(黑色方块)和测试集(红色圆形)的实验Tc与预测Tc

Fig. 7. Experimental Tc and prediction Tc for training set (black square) and test set (red circle)

下载图片 查看所有图片

4 结论

从热力学角度, Tc与约合质量μ的二次多项式关系表明, Tcμ增加而上升, 但拟合结果误差较大。选取六个相对独立的基础描述符, 通过SISSO复杂运算构建得到的四维预测模型显著提升了预测精度。其中,μA/BA/B_NU对Tc影响最大。预测模型的RMSE为0.59 ℃, MaxAE为1.38 ℃, 验证集的MRE为1.00%。描述符的选取(拆分、比值、约合)和修正方式还可以进一步降低预测误差。本课题构建的四维预测模型既可以用于预测其它陶瓷体系的居里温度, 还可以进一步用于构建主动学习框架, 能够快速、简便地预测钙钛矿型压电陶瓷的居里温度。

参考文献

[1] CHENJ, QIY, SHIG, et al.A high temperature piezoelectric ceramic: (1-x)(Bi0.9La0.1)FeO3-xPbTiO3 crystalline solutions. IEEE Trans. Ultrason. Ferroelectr. Freq. Control., 2009, 56(9):1820-1825. 10.1109/TUFFC.2009.1255http://ieeexplore.ieee.org/document/5278429/

[2] CHENJ, JIND, CHENGJ.Impedance spectroscopy studies of 0.7Bi(Fe1-xGax)O3-0.3PbTiO3 high temperature piezoelectric ceramics. J. Alloys Compd., 2013, 580:67-71. 10.1016/j.jallcom.2013.04.076https://linkinghub.elsevier.com/retrieve/pii/S0925838813009596

[3] CHENGJ, YUS, CHENJ, et al.Dielectric and magnetic enhancements in BiFeO3-PbTiO3 solid solutions with La doping. Appl. Phys. Lett., 2006, 89(12):122911. 10.1063/1.2353806http://aip.scitation.org/doi/10.1063/1.2353806

[4] CHENJ, CHENGJ.Enhanced high-field strain and reduced high- temperature dielectric loss in 0.6(Bi0.9La0.1)(Fe1-xTix)O3-0.4PbTiO3 piezoelectric. Ceram. Int., 2015, 41(1):1617-1621. 10.1016/j.ceramint.2014.09.099https://linkinghub.elsevier.com/retrieve/pii/S0272884214014849

[5] 谢颖.ABO3型钙钛矿的相变机理表面稳定性和电子结构的理论研究. 哈尔滨: 黑龙江大学出版社, 2015.

[6] MURPHYK.Machine learning:a probabilistic perspective. Cambridge: The MIT Press, 2012, 58(8):27-71.

[7] PEDREGOSAF, VAROQUAUXG, GRAMFORTA, et al.Scikit- learn: machine learning in python. Journal of Machine Learning Research, 2011, 12(10):2825-2830.

[8] RUPPM, TKATCHENKOA, MÜLLERK, et al.Fast and accurate modeling of molecular atomization energies with machine learning. Phy. Rev. Lett., 2012, 108(5):058301. 10.1103/PhysRevLett.108.058301https://link.aps.org/doi/10.1103/PhysRevLett.108.058301

[9] JORDANM, MITCHELLT.Machine learning: trends, perspectives, and prospects. Science, 2015, 349(6245):255-260. 10.1126/science.aaa841526185243Machine learning addresses the question of how to build computers that improve automatically through experience. It is one of today's most rapidly growing technical fields, lying at the intersection of computer science and statistics, and at the core of artificial intelligence and data science. Recent progress in machine learning has been driven both by the development of new learning algorithms and theory and by the ongoing explosion in the availability of online data and low-cost computation. The adoption of data-intensive machine-learning methods can be found throughout science, technology and commerce, leading to more evidence-based decision-making across many walks of life, including health care, manufacturing, education, financial modeling, policing, and marketing. Copyright © 2015, American Association for the Advancement of Science.

[10] ZHONGM, TRANK, MINY, et al.Accelerated discovery of CO2 electrocatalysts using active machine learning. Nature, 2020, 581(7807):178-183. 10.1038/s41586-020-2242-8https://doi.org/10.1038/s41586-020-2242-8

[11] BATRAR.Accurate machine learning in materials science facilitated by using diverse data sources. Nature, 2021, 589(7843):524-525. 10.1038/d41586-020-03259-4https://doi.org/10.1038/d41586-020-03259-4

[12] RANDHAWAG, HILLK, KARIL.ML-DSP: Machine learning with digital signal processing for ultrafast, accurate, and scalable genome classification at all taxonomic levels. BioMed Central, 2019, 20(1):267-23.

[13] CHENGZ, ZHUE, CHENN.Application of orthogonal expansion to mapping and modelling. Chemometr., 1993, 7(4):243-253. 10.1002/cem.1180070403https://onlinelibrary.wiley.com/doi/10.1002/cem.1180070403

[14] CHENN, LIC, QINP.Chemical pattern recognition applied to materials optimal design and industry optimization. Chin. Sci. Bull., 1997, 42(10):793-799. 10.1007/BF02882484http://link.springer.com/10.1007/BF02882484

[15] CHENN, LUW, CHENR, et al.Chemometric methods applied to industrial optimization and materials optimal design. Chemom. Intel. Lab. Syst., 1999, 45(1):329-333. 10.1016/S0169-7439(98)00139-7https://linkinghub.elsevier.com/retrieve/pii/S0169743998001397

[16] CHENN, ZHUD, WANGW.Intelligent materials processing by hyperspace data mining. Eng. Appl. Artif. Intel., 2000, 13(5):527-532. 10.1016/S0952-1976(00)00032-4https://linkinghub.elsevier.com/retrieve/pii/S0952197600000324

[17] RESTAR.Macroscopic polarization in crystalline dielectrics: the geometric phase approach. Review Modern Physics, 1994, 66(3):899-916. 10.1103/RevModPhys.66.899https://link.aps.org/doi/10.1103/RevModPhys.66.899

[18] ARMIENTOR, KOZINSKYB, FORNARIM, et al.Screening for high-performance piezoelectrics using high-throughput density functional theory. Physics Review B, 2011, 84(1):014103. 10.1103/PhysRevB.84.014103https://link.aps.org/doi/10.1103/PhysRevB.84.014103

[19] PRASANNAV, BENJAMINK, ALPS, et al.Experimental search for high-temperature ferroelectric perovskites guided by two-step machine learning. Nature Communications, 2018, 9:1668. 10.1038/s41467-018-03821-929700297Experimental search for high-temperature ferroelectric perovskites is a challenging task due to the vast chemical space and lack of predictive guidelines. Here, we demonstrate a two-step machine learning approach to guide experiments in search of xBi[Me'Me-y ''((1-y))]O-3-(1 - x) PbTiO3-based perovskites with high ferroelectric Curie temperature. These involve classification learning to screen for compositions in the perovskite structures, and regression coupled to active learning to identify promising perovskites for synthesis and feedback. The problem is challenging because the search space is vast, spanning similar to 61,500 compositions and only 167 are experimentally studied. Furthermore, not every composition can be synthesized in the perovskite phase. In this work, we predict x, y, Me', and Me '' such that the resulting compositions have both high Curie temperature and form in the perovskite structure. Outcomes from both successful and failed experiments then iteratively refine the machine learning models via an active learning loop. Our approach finds six perovskites out of ten compositions synthesized, including three previously unexplored {Me'Me ''} pairs, with 0.2Bi (Fe0.12Co0.88)O-3-0.8PbTiO(3) showing the highest measured Curie temperature of 898 K among them.

[20] EVANM, SUHASY, ILYAG.Prediction of the Curie temperatures of ferroelectric solid solutions using machine learning methods. Computational Materials Science, 2021, 199(7061):110730. 10.1016/j.commatsci.2021.110730https://linkinghub.elsevier.com/retrieve/pii/S0927025621004572

[21] OUYANGR, CURTAROLOS, AHMETCIKE, et al.SISSO: A compressed-sensing method for identifying the best low-dimensional descriptor in an immensity of offered candidates. Phys. Rev. Mater., 2018, 2(8):083802.

[22] OUYANGR, AHMETCIKE, CARBOGNOC, et al.Simultaneous learning of several materials properties from incomplete databases with multi-task SISSO. J. Phys.: Mater., 2019, 2(2):024002. 10.1088/2515-7639/ab077bhttps://iopscience.iop.org/article/10.1088/2515-7639/ab077b

[23] OUYANGR.Exploiting ionic radii for rational design of halide perovskites. Chem. Mater., 2020, 32(1):595-604. 10.1021/acs.chemmater.9b04472https://pubs.acs.org/doi/10.1021/acs.chemmater.9b04472

[24] NINGZ.Dielectric, ferroelectric, piezoelectric and aging properties of BiFeO3-PbTiO3-BaTiO3 high temperature piezoelectric ceramics. Shanghai: Master Thesis, Shanghai University, 2020.

[25] TUT.Fabrication of BF-PT-BT high temperature piezoelectric ceramics and sensors. Shanghai: Master Thesis, Shanghai University, 2017.

[26] CHRISTOPHERJ, CHRISTOPHERS, BRYANR, et al.New tolerance factor to predict the stability of perovskite oxides and halides. Science Advances, 2019, 5(2):eaav0693. 10.1126/sciadv.aav0693https://www.science.org/doi/10.1126/sciadv.aav0693

[27] BUTLERK, FROSTJ, SKELTONJ, et al.Computational materials design of crystalline solids. Che. Soc. Rev., 2016, 45(22):6138-6146.

[28] YUJ, ITOHM.Physics-guided data-mining driven design of room- temperature multiferroic perovskite oxides. Phys. Status Solidi RRL, 2019, 13(6):1900028. 10.1002/pssr.201900028https://onlinelibrary.wiley.com/doi/10.1002/pssr.201900028

[29] UUSIE, MALMJ, IMAMURAN, et al.Characterization of RMnO3 (R=Sc, Y, Dy-Lu): high-pressure synthesized metastable perovskites and their hexagonal precursor phases. Materials Chemistry & Physics, 2008, 112(3):1029-1034.

[30] SCHOBERP, BOERC, SCHWARTEL.Correlation coefficients: appropriate use and interpretation. Anesthesia and Analgesia, 2018, 126(5):1763-1768. 10.1213/ANE.000000000000286429481436Correlation in the broadest sense is a measure of an association between variables. In correlated data, the change in the magnitude of 1 variable is associated with a change in the magnitude of another variable, either in the same (positive correlation) or in the opposite (negative correlation) direction. Most often, the term correlation is used in the context of a linear relationship between 2 continuous variables and expressed as Pearson product-moment correlation. The Pearson correlation coefficient is typically used for jointly normally distributed data (data that follow a bivariate normal distribution). For nonnormally distributed continuous data, for ordinal data, or for data with relevant outliers, a Spearman rank correlation can be used as a measure of a monotonic association. Both correlation coefficients are scaled such that they range from -1 to +1, where 0 indicates that there is no linear or monotonic association, and the relationship gets stronger and ultimately approaches a straight line (Pearson correlation) or a constantly increasing or decreasing curve (Spearman correlation) as the coefficient approaches an absolute value of 1. Hypothesis tests and confidence intervals can be used to address the statistical significance of the results and to estimate the strength of the relationship in the population from which the data were sampled. The aim of this tutorial is to guide researchers and clinicians in the appropriate use and interpretation of correlation coefficients.

[31] YANGX, LIM, SUQ, et al.QSAR studies on pyrrolidine amides derivatives as DPP-IV inhibitors for type 2 diabetes. Medicinal Chemistry Research, 2013, 22(11):5274-5283. 10.1007/s00044-013-0527-2http://link.springer.com/10.1007/s00044-013-0527-2

焦志翔, 贾帆豪, 王永晨, 陈建国, 任伟, 程晋荣. 基于机器学习的BiFeO3-PbTiO3-BaTiO3固溶体居里温度预测[J]. 无机材料学报, 2022, 37(12): 1321. Zhixiang JIAO, Fanhao JIA, Yongchen WANG, Jianguo CHEN, Wei REN, Jinrong CHENG. Curie Temperature Prediction of BiFeO3-PbTiO3-BaTiO3 Solid Solution Based on Machine Learning[J]. Journal of Inorganic Materials, 2022, 37(12): 1321.

引用该论文: TXT   |   EndNote

相关论文

加载中...

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!