Memory efficient neural model improves 3d turbulence prediction accuracy
Scale-Split Neural Operator for Memory- and Data-Efficient 3D Turbulence Prediction
Machine LearningArtificial Intelligence
Summary
Simulating 3D turbulence at high detail is usually slow and needs a lot of computer memory and data. The authors created ScaleSplit-NO, a method that splits this complex problem into two parts: one predicts a rough overall picture and the other fills in the fine details only where needed. This approach makes predicting turbulence faster, uses much less memory, and requires less training data while still being more accurate than previous methods. They tested this on high-resolution datasets and real urban wind simulations, showing significant improvements.
What this means in practice
- •For weather modelers: Speed up urban-scale wind simulations by predicting high-detail turbulence with low memory and data costs.
- •For computational fluid dynamics engineers: Improve accuracy and resource use of turbulence simulations in engineering designs without full high-resolution computations.
Authors
Shaoxiang Qin, Yucheng Zhao, Zongyi Li, Liangzhu Leon Wang, Xiongye Xiao
Abstract
Neural surrogates have emerged as fast alternatives to the numerical simulation of three-dimensional turbulence. However, training them at high resolution remains challenging, since the memory of full-field models grows with the resolution. In addition, full-resolution training data are expensive to simulate and store, and therefore scarce. We introduce ScaleSplit-NO (Scale-Split Neural Operator), which exploits the scale structure of turbulence with two neural operators: a Parent predicts the global coarse field at the next time step, and a Child predicts full-resolution local patches conditioned on this prediction. Neither model operates on the full-resolution field. The Child is pretrained alone and then attached to the Parent's coarse prediction through zero-initialized connections. On two complex high-resolution turbulence benchmarks, ScaleSplit-NO surpasses all competing baselines in both prediction accuracy and data efficiency. On the higher-resolution dataset JHTDB256 ($256^3$), its normalized mean squared error (NMSE) is 53% lower than that of the strongest baseline, and its training memory is 79% lower than that of the most memory-efficient baseline. We further demonstrate its effectiveness for urban wind prediction in a real district of Montreal on a $500\times150\times500$ grid, reducing one-step NMSE by 65.8% relative to the baseline. Moreover, swapping in a Parent trained on additional coarse fields improves prediction without retraining the Child, providing further accuracy gains at a small storage cost.