IDP-PGFE: An Interpretable Disruption Predictor based on Physics-Guided Feature Extraction

• DOI: 10.48550/arxiv.2208.13197
• Publication Date: 2022-01-01
• Authors: Chengshuo Shen; Wei Zheng; Yonghua Ding; Xinkun Ai; Fengming Xue; Yu Zhong; Nengchao Wang; Gao Li; Zhipeng Chen; Zhoujun Yang; Zhongyong Chen; Yuan Pan; J-Text Team
• Keywords: Interpretability; Intuition

Disruption prediction has made rapid progress in recent years, especially machine learning (ML)-based methods. Understanding why a predictor makes certain can be as crucial the prediction's accuracy for future tokamak disruption predictors. The purpose of most predictors is or cross-machine capability. However, if model interpreted, it tell samples are classified precursors. This allows us to types incoming and gives insight into mechanism disruption. paper designs called Interpretable Predictor based On Physics-guided feature extraction (IDP-PGFE) on J-TEXT. performance effectively improved by extracting physics-guided features. A high-performance required ensure validity interpretation results. interpretability study IDP-PGFE provides an understanding J-TEXT generally consistent with existing comprehension been applied due continuously increasing density towards limit experiments time evolution PGFE features contribution demonstrates that application ECRH triggers radiation-caused disruption, which lowers at While RMP indeed raises guides intuition physical mechanisms RMPs affect not only MHD instabilities but also radiation profile, delays