本文采用机器学习算法对晶体的生成焓的预测进行了研究。利用由多个神经层组成的深度神经网络(DNN)模型学习数据间的关系，对深度学习模型在材料结生成焓的预测做了深入的研究和讨论。通过对开放量子材料数据库(OQMD)中275,778种化合物的生成焓参数进行学习，建立了深度学习多层全连接网络，并用来预测未知晶体材料的生成焓。优化后预测模型的精度达到了0.075 eV/atom，达到了量子力学软件的计算精度。
The prediction of enthalpy of formation of the crystal is studied by the machine learning algo-rithm. Based on the relationship between the learning data of the deep neural network (DNN) model composed of multiple neural layers, the prediction of enthalpy of formation of the deep learning model in the material junction is studied and discussed in depth. By learning the en-thalpy of formation parameters of 275,778 compounds in the open quantum materials database (OQMD), a deep learning multi-layer fully connected network was established to predict the en-thalpy of formation of unknown crystal materials. The accuracy of the optimized prediction model reached 0.075 eV/atom, which reached the computational accuracy of the quantum me-chanics software.