Configuration method of soil moisture reconstruction model under coupling of physical constraint and deep learning

• Main Authors: MIAO LINGUANG, CHEN JINGJING, CHAO TIENIEH, MENG LINGKUI, WEI ZUSHUAI
• Format: Patent
• Language: Chinese; English
• Published: 29.12.2023
• Subjects: CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention provides a configuration method of a soil moisture reconstruction model under coupling of physical constraint and deep learning, which is used for dividing dynamic change of soil moisture into a physical process of two stages and specifically fusing the dynamic change into configuration content of a space-time seamless soil moisture reconstruction model in a loss function penalty term form. In this way, the deep learning model solution space can be constrained based on a physical mechanism, the inference interpretation ability for unknown data is improved, the space-time seamless soil moisture reconstruction model obtained through training can complete reconstruction processing with high precision in the face of soil moisture data of SMAP, the problem of information loss caused by satellite scanning gaps is solved, and the real-time performance of the soil moisture reconstruction model is improved. And the method has better practical value. 本申请提供了涉及一种物理约束与深度学习耦合下土壤水分重建模型的配置方法，用于将土壤水分的动态变化划分为两个阶段