On the transfer of adaptive predictors between different devices for both mitigation and prevention of disruptions

回到主页

https://iopscience.iop.org/article/10.1088/1741-4326/ab77a6
This paper talks about transfer learning beteen devices of Jet and AUG。Jet has used the Random Forest Model trained by data of Jet to predict disruption on AUG and vice versa.Many of these weak classifiers were trained and then combined to obtain more stable and performing results。
4

diagnostic difference beteen of Jet and AUG

locked mode 4

internal inductance and time constant 6

diagnostic selection

This paper doesn't show all diagnostics they used. It just mentions internal inductance、 locked mode and total radiation.
6

AUG and JET databases and statistics

8

Results: transfer from JET to AUG

Mitigation on AUG

The success rate is 87.66% (135/154) and the average warning time 22.3 ms. With regard to the false alarm rate, the final performance is 5.70% (31/538).

Toward prevention on AUG

8

Results: transfer from AUG to JET

Transfer to JET for mitigation

8

Transfer to JET for prevention

8

Inspiration

  1. For transfer learning, we can train many weak classifiers and then find a decision function to ensemble the result of those weak classifiers into the final result.
  2. Before predicting disruption cross the device, we should compare difference of each device and unify the different forms of diagnosis.

本文章使用limfx的vscode插件快速发布