A Physics-Informed Neural Network Source Iteration Approach for Improved Accuracy and Convergence in Neutron Diffusion Problems: Dataset and Benchmark Results

Research on the PINN framework built on PyTorch, covering three typical testing scenarios: (1) the five region problem of smooth transition; (2) TWIGL benchmark questions (including two sets of results before and after section correction); (3) XPINN Three Partition Example.

The dataset provides high-resolution images, training logs, and key model outputs, which can be used for validation research of methods such as PINN, RAR, dynamic weighting, XPINN, etc.