Cross-tokamak disruption prediction based on domain adaptation

• DOI: 10.1088/1741-4326/ad3e12
• Publication Date: 2024-04-12T22:23:16Z
• Authors: Chengshuo Shen; Wei Zheng; Bihao Guo; Yonghua Ding; Dalong Chen; Xinkun Ai; Fengming Xue; Yu Zhong; Nengchao Wang; Biao Shen; Binjia Xiao; Zhongyong Chen; Yuan Pan
• Keywords: Feature (linguistics)

Abstract The high acquisition cost and the significant demand for disruptive discharges data-driven disruption prediction models in future tokamaks pose an inherent contradiction research. In this paper, we demonstrated a novel approach to predict tokamak using only few based on domain adaptation (DA). aims by finding feature space that is universal all tokamaks. first step use existing understanding of physics extract physics-guided features from diagnostic signals each tokamak, called extraction (PGFE). second align data (target domain) large amount (source DA algorithm CORrelation ALignment (CORAL). It attempt at applying cross-tokamak task. PGFE has been successfully applied J-TEXT with excellent performance. can also reduce volume requirements due extracting less device-specific features, thereby establishing solid foundation prediction. We have further improved CORAL supervised (S-CORAL) enhance its appropriateness alignment To simulate case, selected as EAST which gap ranges plasma parameters. utilization S-CORAL improves performance tokamak. Through interpretable analysis, discovered learned knowledge model through exhibits more similarities trained volumes This provides light, data-required ways aligning small