DNA: A Double-Helix Hierarchical Deep Interaction Network Architecture for Industrial Digital Twin
Organizations: Department of Control Science and Engineering, Zhejiang University, Hangzhou, China · Risk Department, Binance
Abstract
High-fidelity modeling is fundamental to industrial digital twin, yet a trade-off persists between the physical transparency of mechanistic models and the superior predictive power of data-driven approaches. Existing hybrid methods often fail to bridge this gap, suffering from shallow integration and rigid one-shot modeling strategies that struggle with complex industrial systems. To address these issues, we propose a Double-Helix Hierarchical Deep Interaction Network Architecture (DNA). Inspired by the double-helix structure of biological DNA, our architecture intertwines mechanistic and data-driven modeling strands through two novel interaction strategies, forming a symbiotic architecture in which the two strands mutually enhance predictive capability. Furthermore, the hierarchical structure of DNA allows it to model complex systems in a multi-stage manner through two tailored deep learning frameworks. We evaluate the proposed approach on two distinct industrial systems, where DNA achieves reductions in MAE of 3.704% and 3.175%, respectively, compared to the strongest baseline. These results demonstrate the effectiveness of DNA and suggest its potential to provide new insights for the development of industrial digital twins.
Figures & tables
| Category | DNA | Tree-based | Transformer-based | Graph-based | Others | |||||||||||||||||||||||
| Model | Ours | LightGBM | Transformer | FlowformerMH | iTransformer | MPFormer | GATv2 | DGDL | TGCNS | RaDA | TimesNet | TimeXer | Umixer | |||||||||||||||
| Metric | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | ||
| Stage I | Cp.g | MM | 49.58 | 50.72 | 49.58 | 50.72 | 49.58 | 50.72 | 49.58 | 50.72 | 49.58 | 50.72 | 49.58 | 50.72 | 49.58 | 50.72 | 49.58 | 50.72 | 49.58 | 50.72 | 49.58 | 50.72 | 49.58 | 50.72 | 49.58 | 50.72 | 49.58 | 50.72 |
| DD | 0.763 | 1.073 | 0.913 | 1.205 | 1.754 | 2.013 | 1.467 | 1.965 | 1.037 | 1.312 | 0.795 | 1.073 | 0.851 | 1.094 | 1.387 | 1.736 | 0.832 | 1.065 | 1.300 | 1.619 | 0.794 | 1.060 | 0.807 | 1.068 | 0.843 | 1.100 | ||
| Cor | 0.831 | 1.067 | 1.253 | 1.943 | 0.780 | 1.111 | 1.130 | 1.747 | 0.812 | 1.120 | 0.770 | 1.051 | 0.799 | 1.143 | 0.823 | 1.101 | 0.767 | 1.073 | 0.792 | 1.138 | 0.859 | 1.130 | 0.792 | 1.069 | ||||
| Avg | 0.797 | 1.070 | 1.503 | 1.978 | 1.123 | 1.538 | 1.083 | 1.530 | 0.804 | 1.096 | 0.811 | 1.073 | 1.093 | 1.440 | 0.827 | 1.083 | 1.034 | 1.346 | 0.793 | 1.099 | 0.833 | 1.099 | 0.817 | 1.084 | ||||
| Stage II | HL | P | ||
|---|---|---|---|---|
| MAE | RMSE | MAE | RMSE | |
| w/o GCN-E | 1.612 | 2.726 | 4.509 | 7.641 |
| w/o Gating | 1.757 | 2.891 | 8.858 | 12.48 |
| Ours | 1.556 | 2.677 | 3.017 | 5.424 |
| Stage I | Cp.g | Cp.o | Cp.w | Rho.o | Rho.l | Vis.w | Vis.o | Vis.l | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | ||||
| Ours | DD | 0.763 | 1.073 | 1.704 | 2.335 | 0.724 | 1.017 | 0.793 | 1.367 | 2.434 | 3.437 | 2.191 | 3.125 | 28.27 | 46.84 | 31.82 | 46.65 | ||
| Cor | 0.831 | 1.067 | 1.895 | 2.474 | 1.219 | 1.611 | 1.504 | 2.397 | 6.583 | 7.946 | 2.373 | 3.286 | 36.98 | 53.48 | 44.00 | 60.20 | |||
| Avg | 0.797 | 1.070 | 1.799 | 2.404 | 0.972 | 1.314 | 1.148 | 1.882 | 4.509 | 5.692 | 2.282 | 3.205 | 32.62 | 50.16 | 37.91 | 53.43 | |||
| MASA MHSA | DD | 0.963 | 1.231 | 1.702 | 2.372 | 1.333 | 1.532 | 0.802 | 1.394 | 2.144 | 3.451 | 2.252 | 3.179 | 28.31 | 46.82 | 32.10 | 47.19 | ||
| Cor | 0.787 | 1.091 | 1.962 | 2.543 | 1.167 | 1.542 | 1.575 | 2.382 | 6.726 | 8.151 | 2.520 | 3.311 | 59.74 | 85.06 | 43.72 | 60.46 | |||
| Model | DNA | Transformer | FlowformerMH | iTransformer | MPFormer | GATv2 | DGDL | TGCNS | RaDA | TimesNet | TimeXer | Umixer | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Metric | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | ||
| HL | MM | 8.513 | 10.91 | 8.513 | 10.91 | 8.513 | 10.91 | 8.513 | 10.91 | 8.513 | 10.91 | 8.513 | 10.91 | 8.513 | 10.91 | 8.513 | 10.91 | 8.513 | 10.91 | 8.513 | 10.91 | 8.513 | 10.91 | 8.513 | 10.91 | |
| Cor | 5.417 | 7.199 | 5.445 | 7.175 | 5.449 | 7.180 | 5.424 | 7.129 | 5.418 | 7.129 | 5.419 | 7.133 | 5.414 | 7.131 | 5.416 | 7.136 | 5.413 | 7.126 | 5.426 | 7.142 | 5.419 | 7.117 | 5.440 | 7.148 | ||
| P | MM | 124.2 | 214.5 | 124.2 | 214.5 | 124.2 | 214.5 | 124.2 | 214.5 | 124.2 | 214.5 | 124.2 | 214.5 | 124.2 | 214.5 | 124.2 | 214.5 | 124.2 | 214.5 | 124.2 | 214.5 | 124.2 | 214.5 | 124.2 | 214.5 | |
| Cor | 77.55 | 131.2 | 92.05 | 180.1 | 90.22 | 171.4 | 93.49 | 184.8 | 90.47 | 177.1 | 92.03 | 179.4 | 91.91 | 179.3 | 91.83 | 179.3 | 91.69 | 178.9 | 90.90 | 177.6 | 94.94 | 187.0 | 90.62 | 176.7 | ||
Appendix figures & tables11 assets
Supplementary material from the paper’s appendix.
Appendix
| Modeled object | Unit | ||||
| Stage I | Cp.g | , | , | ||
| Cp.o | , | ||||
| Cp.w | , | ||||
| Rho.o | , | ||||
| Rho.l | , | ||||
| Vis.w | , | ||||
| Stage | Dataset | Batch Size | Dim | lr | Max Training Epochs | Early Stopping at Epoch |
| Stage I | Cp.g | 256 | 32 | 1e-4 | 2000 | 1733 |
| Cp.o | 256 | 32 | 1e-4 | 2000 | 1140 | |
| Cp.w | 256 | 32 | 5e-4 | 2000 | 502 | |
| Rho.o | 256 | 32 | 1e-4 | 2000 | 1236 | |
| Rho.l | 256 | 32 | 5e-4 | 2000 | 2000 | |
| Vis.w | 256 | 32 | 1e-4 | 2000 | 1636 |
| Dataset | Correction Function | |||
| Cp.g | 2 | ✗ | ✓ | |
| Cp.o | 2 | ✗ | ✓ | |
| Cp.w | 0 | ✓ | ✗ | |
| Rho.o | 2 | ✓ | ✓ | |
| Rho.l | 2 | ✓ | ✗ | |
| Vis.w | 2 | ✓ | ✗ |
| Models | Parameters | FLOPs | Inference Time (ms) |
|---|---|---|---|
| DNA | 8726 | 67540 | 32.85 |
| Transformer | 65476 | 213984 | 12.38 |
| FlowFormer | 16253 | 243241 | 12.95 |
| iTransformer | 135460 | 89264 | 19.28 |
| MPFormer | 77444 | 210288 | 81.78 |
| GATV2 | 32722 | 134752 | 14.70 |
| Dataset | FLOPs(DNA with MASA) | FLOPs(DNA with MHSA) | FLOPs Reduction (%) |
|---|---|---|---|
| Cp.g | 67540 | 79892 | 15.46 |
| Cp.o | 70740 | 79892 | 11.46 |
| Cp.w | 67412 | 79892 | 15.62 |
| Rho.o | 45028 | 51620 | 12.77 |
| Rho.l | 72560 | 87280 | 16.87 |
| Vis.w | 53760 | 58496 | 8.10 |
Digital Twins
Digital twins are virtual models synchronized with real systems to support monitoring, prediction, simulation, and decision-making.
See all 284 papers