Scholarlink

<p>This study aims to overcome the high computational cost and time-consuming challenges faced by traditional ABX₃ perovskite crystal structure prediction methods, and innovatively proposes a prediction model that integrates random forest and a convolutional neural network. This study employs synthetic minority oversampling techniques to address data imbalance issues and utilizes feature selection and multi-layer convolution pooling operations to achieve deep automatic learning of perovskite crystal structure data. The experimental results showed that the model had excellent performance. Its prediction accuracy could reach up to 97.41%, and it performed well in training time, with a minimum of 21.3&#xa0;min. The model also proved to be memory-efficient, with peak training memory consumption controlled at a relatively low level, as low as 195.76&#xa0;MB. Moreover, when facing different tolerance factors, the prediction error range of the model remained stable between 3.12% and 4.87%, with excellent stability and consistency. Compared with models such as backpropagation networks, long short-term memory networks, and machine learning atomic potentials, the proposed model exhibited significant advantages in prediction accuracy, training time, and computational resource consumption. This research achievement is of great significance, opening up an efficient and accurate new path for predicting the crystal structure of ABX₃ perovskite. This helps researchers quickly screen and design target materials during the development of new materials, accelerating the discovery and development process of new materials.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Prediction of ABX3 perovskite crystal structure based on random forest algorithm

  • 期刊论文

  • 发表日期：

    2025年11月20日

查看全文
  • Xueshuang Deng,
  • Jiaojiao Chen,
  • Qi Luo

摘要

This study aims to overcome the high computational cost and time-consuming challenges faced by traditional ABX₃ perovskite crystal structure prediction methods, and innovatively proposes a prediction model that integrates random forest and a convolutional neural network. This study employs synthetic minority oversampling techniques to address data imbalance issues and utilizes feature selection and multi-layer convolution pooling operations to achieve deep automatic learning of perovskite crystal structure data. The experimental results showed that the model had excellent performance. Its prediction accuracy could reach up to 97.41%, and it performed well in training time, with a minimum of 21.3 min. The model also proved to be memory-efficient, with peak training memory consumption controlled at a relatively low level, as low as 195.76 MB. Moreover, when facing different tolerance factors, the prediction error range of the model remained stable between 3.12% and 4.87%, with excellent stability and consistency. Compared with models such as backpropagation networks, long short-term memory networks, and machine learning atomic potentials, the proposed model exhibited significant advantages in prediction accuracy, training time, and computational resource consumption. This research achievement is of great significance, opening up an efficient and accurate new path for predicting the crystal structure of ABX₃ perovskite. This helps researchers quickly screen and design target materials during the development of new materials, accelerating the discovery and development process of new materials.