基于多物理信息神经网络的心墙堆石坝多源数据同化研究OA

Thorough perception of hydraulic structures is essential for building their digital twins and promoting high-quality water resources development.With the advancement of dam safety monitoring and the establishment of inte-grated"sky-space-ground-water-structure based"sensing networks,effectively assimilating multi-source data to achieve full-domain perception has become a key research focus.This study proposes a Multiphysics-Informed Neural Network(MPINN)data assimilation framework,which embeds stress-seepage coupling theory and random field parameters into neural networks to integrate physical consistency with data-driven flexibility.Taking the Lianghekou core wall rockfill dam as an example,the MPINN was used to assimilate monitoring data of pore water pressure,earth pressure in the gravelly soil core wall,as well as detection data of permeability coefficient and compression modulus to reconstruct the full-field pore water pressure and earth pressure distributions within the core wall from sparse data.Comparative experiments showed that the MPINN outperforms other methods in both prediction accuracy and robust-ness,verifying the effectiveness of the data assimilation.This provides a new pathway for achieving thorough percep-tion and offers technical support for the development of digital twin projects and intelligent dams.