(1. 上海大學(xué) 材料基因組工程研究院，上海 200444；
2. 上海大學(xué) 計(jì)算工程與科學(xué)學(xué)院，上海 200444；
3. 上海大學(xué) 理學(xué)院，上海 200444)
摘 要: 鈣鈦礦材料由于在各領(lǐng)域具有廣泛的應(yīng)用前景而備受材料學(xué)家的關(guān)注，對(duì)其各種物理化學(xué)性能的研究一直是材料領(lǐng)域研究的熱點(diǎn)。本文建立隨機(jī)森林(Random forest，RF)、嶺回歸(Ridge regression，RR)、以及基于徑向基核函數(shù)和線性核函數(shù)的支持向量回歸(Support vector regression，SVR)等4種機(jī)器學(xué)習(xí)算法的預(yù)測(cè)模型，對(duì)鈣鈦礦材料數(shù)據(jù)集中的密度、形成能、帶隙、晶體體積等4種性能參數(shù)進(jìn)行預(yù)測(cè)。結(jié)果表明：RF方法可以對(duì)鈣鈦礦材料的密度、帶隙性能進(jìn)行有效預(yù)測(cè)；RR方法可以實(shí)現(xiàn)對(duì)密度性能的預(yù)測(cè)；線性核函數(shù)的SVR方法可以實(shí)現(xiàn)對(duì)形成能性能的預(yù)測(cè)。該研究表明，不同的機(jī)器學(xué)習(xí)算法對(duì)數(shù)據(jù)樣本分布的敏感程度不同，因此針對(duì)不同的性能參數(shù)預(yù)測(cè)需要選擇不同方法。
關(guān)鍵字: 鈣鈦礦材料；機(jī)器學(xué)習(xí)；性能預(yù)測(cè)；算法選擇
（1. College of Sciences, Shanghai University, Shanghai 200444, China;
2. School of Computer Engineering and Science, Shanghai University, Shanghai 200444, China;
3. School of Materials Science and Engineering, Shanghai University, Shanghai 200444, China）
Abstract:Due to the potential application in various fields, there is great opportunity for further research into the basic physics and chemistry around perovskites. In this work, four machine learning algorithms prediction models have been built. They are random forest(RF), ridge regression(RR), and support vector regression(SVR) based on the radial basis kernel function and linear kernel function. They are used to predict the density, formation energy, band gap, and the crystal volume of the perovskite materials. The experimental results show that the RF method can effectively predict the density and band gap of perovskite materials. The RR method can realize the prediction of density performance. The SVR method of linear kernel function can realize the prediction of the performance. This study shows that different machine learning algorithms have different sensitivity to the distribution of data set samples, so different methods should be selected to predict different performance parameters.
Key words: perovskite; machine learning; performance prediction; algorithm selection
