Spatiotemporal Field Generation Based on Hybrid Mamba-Transformer with Physics-informed Fine-tuning

TL;DR

This paper introduces HMT-PF, a hybrid Mamba-Transformer model with physics-informed fine-tuning for generating accurate spatiotemporal physical fields, effectively reducing physical discrepancies through self-supervised learning.

Contribution

The paper presents a novel hybrid Mamba-Transformer architecture with a physics-informed fine-tuning mechanism for improved spatiotemporal field generation.

Findings

The model achieves good performance in generating physical fields.

Physics-informed fine-tuning reduces physical equation discrepancies.

A new MSE-R evaluation method assesses accuracy and realism.

Abstract

This research confronts the challenge of substantial physical equation discrepancies encountered in the generation of spatiotemporal physical fields through data-driven trained models. A spatiotemporal physical field generation model, named HMT-PF, is developed based on the hybrid Mamba-Transformer architecture, incorporating unstructured grid information as input. A fine-tuning block, enhanced with physical information, is introduced to effectively reduce the physical equation discrepancies. The physical equation residuals are computed through a point query mechanism for efficient gradient evaluation, then encoded into latent space for refinement. The fine-tuning process employs a self-supervised learning approach to achieve physical consistency while maintaining essential field characteristics. Results show that the hybrid Mamba-Transformer model achieves good performance in generating spatiotemporal fields, while the physics-informed fine-tuning mechanism further reduces significant physical errors effectively. A MSE-R evaluation method is developed to assess the accuracy and realism of physical field generation.