Lattice Boltzmann Fluid Solver with Symmetric, Conservative and Equilibrium-Preserving Neural Collision Operator

	Lattice Boltzmann Fluid Solver with Symmetric, Conservative and Equilibrium-Preserving Neural Collision Operator
	Fu, Xinyi; Bai, Kai; Wang, Yang; Liu, Xiaopei
	2025
发表期刊	IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS (IF:10.2[JCR-2023],10.4[5-Year])
ISSN	2162-237X
EISSN	2162-2388
发表状态	待投递
摘要	Neural fluid solvers have been garnering significant interest for their potential to accelerate fluid simulations and create surrogate models from data. However, general neural PDE solvers often face challenges in strictly enforcing physical properties such as symmetry, conservation and equilibrium preservation, which can lead to insufficient generalizability. In our work, we introduce a neural lattice Boltzmann solver with a carefully designed neural collision operator that unconditionally satisfies these essential properties. Our method demonstrates better generalizability and outperforms previous approaches, achieving twice the duration of high vorticity correlation on a down-sampled decaying turbulence dataset.
关键词	lattice Boltzmann method neural fluid solver structure-preserving learning constrained neural network
收录类别	SCIE
文献类型	期刊论文
条目标识符	https://kms.shanghaitech.edu.cn/handle/2MSLDSTB/430465
专题	信息科学与技术学院_硕士生信息科学与技术学院_PI研究组_刘晓培组信息科学与技术学院_PI研究组_汪阳组
通讯作者	Liu, Xiaopei
作者单位	School of Information Science and Technology, ShanghaiTech University
第一作者单位	信息科学与技术学院
通讯作者单位	信息科学与技术学院
第一作者的第一单位	信息科学与技术学院
推荐引用方式 GB/T 7714	Fu, Xinyi,Bai, Kai,Wang, Yang,et al. Lattice Boltzmann Fluid Solver with Symmetric, Conservative and Equilibrium-Preserving Neural Collision Operator[J]. IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS,2025.
APA	Fu, Xinyi,Bai, Kai,Wang, Yang,&Liu, Xiaopei.(2025).Lattice Boltzmann Fluid Solver with Symmetric, Conservative and Equilibrium-Preserving Neural Collision Operator.IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS.
MLA	Fu, Xinyi,et al."Lattice Boltzmann Fluid Solver with Symmetric, Conservative and Equilibrium-Preserving Neural Collision Operator".IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS (2025).