Physics-informed Partitioned Coupled Neural Operator for Complex Networks

Weidong Wu,Yong Zhang,Lili Hao,Yang Chen,Xiaoyan Sun,Dunwei Gong

2024-10-28

Abstract:Physics-Informed Neural Operators provide efficient, high-fidelity simulations for systems governed by partial differential equations (PDEs). However, most existing studies focus only on multi-scale, multi-physics systems within a single spatial region, neglecting the case with multiple interconnected sub-regions, such as gas and thermal systems. To address this, this paper proposes a Physics-Informed Partitioned Coupled Neural Operator (PCNO) to enhance the simulation performance of such networks. Compared to the existing Fourier Neural Operator (FNO), this method designs a joint convolution operator within the Fourier layer, enabling global integration capturing all sub-regions. Additionally, grid alignment layers are introduced outside the Fourier layer to help the joint convolution operator accurately learn the coupling relationship between sub-regions in the frequency domain. Experiments on gas networks demonstrate that the proposed operator not only accurately simulates complex systems but also shows good generalization and low model complexity.

Machine Learning,Computational Engineering, Finance, and Science,Computational Physics

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the challenges faced by existing neural operator methods in dealing with multi - domain complex network systems in complex network systems, especially those containing multiple coupled sub - regions. Specifically, most of the existing research mainly focuses on multi - scale and multi - physical field systems occurring within a single spatial region, while ignoring the situation with multiple interconnected sub - regions, such as natural gas and heating systems. To address this limitation, this paper proposes a physics - informed partition - coupled neural operator (PCNO) aiming to enhance the simulation performance of such networks. PCNO overcomes the deficiencies of existing methods in dealing with multi - domain complex networks by introducing a joint convolution operator to learn the coupling relationships between multiple sub - regions and achieve global integration. In addition, in order to ensure that the joint convolution operator can accurately calculate the global integral in the frequency domain, a grid - alignment layer is proposed to adapt to the sizes of different sub - regions and the non - periodicity of input functions. These improvements enable PCNO not only to accurately simulate complex systems at different resolutions, but also to exhibit good generalization ability and low model complexity.

Physics-informed Partitioned Coupled Neural Operator for Complex Networks

Physics-informed Discretization-independent Deep Compositional Operator Network

Physics-Informed Deep Neural Operator Networks

Physics-Informed Neural Operator for Learning Partial Differential Equations

Enhancing Solutions for Complex PDEs: Introducing Complementary Convolution and Equivariant Attention in Fourier Neural Operators

Learning Only on Boundaries: A Physics-Informed Neural Operator for Solving Parametric Partial Differential Equations in Complex Geometries

Differential Operator Approximation Based Tightly Coupled Multiphysics Solver Using Cascaded Fourier Network

CoNO: Complex Neural Operator for Continous Dynamical Physical Systems

Model-Parallel Fourier Neural Operators as Learned Surrogates for Large-Scale Parametric PDEs

Physics-embedded Fourier Neural Network for Partial Differential Equations

Separable Operator Networks

Physics-Guided, Physics-Informed, and Physics-Encoded Neural Networks and Operators in Scientific Computing: Fluid and Solid Mechanics

FB-HyDON: Parameter-Efficient Physics-Informed Operator Learning of Complex PDEs via Hypernetwork and Finite Basis Domain Decomposition

Physics-enhanced Neural Operator for Simulating Turbulent Transport

Partition of Unity Physics-Informed Neural Networks (POU-PINNs): An Unsupervised Framework for Physics-Informed Domain Decomposition and Mixtures of Experts

Coupled Multiphysics Solver for Irregular Regions Based on Graph Neural Network

Coupled Multiwavelet Neural Operator Learning for Coupled Partial Differential Equations

Diffeomorphism Neural Operator for various domains and parameters of partial differential equations

Physics-Informed Geometry-Aware Neural Operator

Learning the solution operator of parametric partial differential equations with physics-informed DeepONets