Research On Band Gap Prediction Of Double Perovskite Based On Support Vector Machine

Double perovskite solar cell is a research direction with great potential for the field of photovoltaic power generation.This kind of cell has crucial advantages for photovoltaic power generation applications,such as low cost,slow decline from the conversion efficiency,and strong stability.One of the key points for the further development of this kind of cell performance is to explore double perovskite semiconductor materials with more suitable narrow band gap.With the development of artificial intelligence and material databases,the application of machine learning algorithms can achieve the fast and efficient prediction and screening of new material properties.The double perovskite band gap dataset is a typical unbalanced dataset,in which the data onto materials whose band gaps meet the needs of optoelectronic applications account for less than 10%.In the case of extremely unbalanced data,it is difficult for traditional machine learning methods to effectively screen double perovskite materials with potential for optoelectronic application.In this paper,the unbalanced machine learning method is applied to predict and screen the double perovskite materials with photoelectric application potential.The research contents of this paper are as follows:First,for the high data imbalance problem in the band gap dataset of double perovskites,this paper proposes an ISO-SMOTE oversampling algorithm based on ISODATA clustering.The algorithm is used to rebalance the double perovskite bandgap dataset.The ISO-SMOTE algorithm increases the processing of noise and ignores data distribution and other problems,and is an improved algorithm for the shortcomings of the SMOTE oversampling algorithm.The experimental results of public datasets show that the ISO-SMOTE algorithm can improve the model’s ability to distinguish minority samples,which is helpful for achieving better classification results.Secondly,a model for predicting the band gap of double perovskites is established by applying the cost-sensitive support vector machine algorithm.The model can fully consider the imbalance of misclassification cost in imbalanced data sets,and further solve the problem of imbalanced data classification.The experimental result of the double perovskite band gap dataset shows that the algorithm has better performance in multiple evaluation indicators,and has certain advantages in the double perovskite band gap prediction problem.Finally,based on the ISO-SMOTE algorithm and the cost-sensitive support vector machine algorithm described above,a prediction and screening system for the double perovskite band gap is built,and the band gap of the double perovskite material is predicted and screened.The result shows that the application of this screening system can screen out 269 materials most likely to have excellent properties from 16,000 potential double perovskite materials.This approach greatly narrows the screening range of materials,and provides a guiding direction for the exploration of double perovskite materials with potential for optoelectronic applications.