Initialization-enhanced Physics-Informed Neural Network with Domain Decomposition (IDPINN)

Physics-Informed Neural Network (PINN) has emerged as a promising approach to tackle both forward and inverse problems associated with Partial Differential Equations (PDEs) [1]. PINNs incorporate the governing equation into the neural network's architecture at its core, augmenting the loss function with a residual term derived from the equation. This structural adjustment penalizes deviations from the equation, narrowing the range of viable solutions. Consequently, it transforms the process of determining PDE solutions into an optimization task focused on minimizing the loss function. PINNs have demonstrated adaptability and effectiveness across various domains and interdisciplinary fields [2, 3, 4, 5].