The Properties Of Layered Perovskite Thermoelectric Material（SrO）_m（SrTiO₃）_n Were Studied By Machine Learning Method

With the development of thermoelectric materials,the main goal is to find n-type oxide thermoelectric materials with excellent thermoelectric properties.Mass of data generated by the traditional materials research methods through experiments and simulations lays the foundation for the new materials research paradigm.The "big data"generated by these experiments and simulations provides an unprecedented opportunity for the application of data-driven technology in this field.In this paper,the Pymatgen(Python Materials Genomics)is used to obtain various data related to the material performance in the Materials Project(high-throughput material calculation and data sharing platform).After taking into account the structural difference and atomic number difference of material data,a set of appropriate material descriptive parameters(descriptors)was obtained from OQMD database to construct the data set under the condition that the eigenvalue dimensions of the data were consistent.The classification algorithm in machine learning is used to analyze the acquired material data based on available data,and determine that SrTiO3 has the potential to become high-temperature oxide thermoelectric material.The doping modification of materials has always been an important topic in the research of materials.As(SrO)m(SrTiO3)n material is a layered perovskite material,its structure is easy to be adjusted,which can be used for the design of new materials.On the premise that m+n=20,by changing the number of SrO atomic layers in(SrO)m(SrTiO3)n and the position of the SrO layer in the cell,a total of 4496(SrO)m(SrTiO3)n supercells with different structures were designed.Due to the difference in the number of SrO atomic layer and SrTiO3 atomic layer in(SrO)m(SrTiO3)n,as well as the difference in the position of SrO layer and SrTiO3 layer in the cell,the performance of different structures(SrO)m(SrTiO3)n is obviously different and shows a certain variation trend.Using the regression prediction algorithm in machine learning to predict the heat capacity of different structure(SrO)m(SrTiO3)n materials.By comparing the performance of different algorithms,among many algorithms,the strong learner can perform well in the case of small data volume,but it needs more training time than the weak learner,while the neural network can achieve very high accuracy with less training time by adjusting the network structure.As the trained machine learning model can achieve a high accuracy,it can almost replace the traditional computational simulation.It shows that applying the machine learning algorithm model to the research of thermoelectric materials can effectively shorten the research and development cycle.Nowadays,the structure of materials is becoming more and more complicated,and the data of materials are getting larger and larger.This new research paradigm is very suitable for material research.