Interpretability analysis and real-time prediction of locked mode-induced disruptions in EAST

Abstract Locked mode-induced disruptions (LMiDs) are investigated using an interpretable machine learning approach in EAST. The experimental data of LM come from error field penetration experiments conducted between 2015 and 2022. Decision tree ensembles, known for their strong interpretability and accuracy, were selected as the machine learning algorithm to analyze and predict LMiDs. Two prediction models were developed: an interpretable predictor and a real-time predictor. The interpretable predictor, which employs physics-guided feature extraction, achieved exceptional performance on the test set (AUC = 0.997). This model identified the distance between the outer boundary of the magnetic island and the last closed surface of the plasma, denoted as <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mrow> <mml:msub> <mml:mi>d</mml:mi> <mml:mrow> <mml:mi>edge</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> </mml:math> , as the most important feature for LMiDs. Other important features include stored energy <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mrow> <mml:msub> <mml:mi>W</mml:mi> <mml:mrow> <mml:mi>MHD</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> </mml:math> , radial force <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mrow> <mml:msub> <mml:mi>V</mml:mi> <mml:mrow> <mml:mi>loop</mml:mi> </mml:mrow> </mml:msub> <mml:mo>×</mml:mo> <mml:msub> <mml:mi>B</mml:mi> <mml:mrow> <mml:mi mathvariant="normal">p</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> </mml:math> , total heating power <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mrow> <mml:msub> <mml:mi>P</mml:mi> <mml:mrow> <mml:mi>total</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> </mml:math> , toroidal magnetic fields <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mrow> <mml:msub> <mml:mi>B</mml:mi> <mml:mrow> <mml:mi mathvariant="normal">t</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> </mml:math> , internal inductance <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mrow> <mml:msub> <mml:mi>l</mml:mi> <mml:mrow> <mml:mi mathvariant="normal">i</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> </mml:math> and electron temperature <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mrow> <mml:msub> <mml:mi>T</mml:mi> <mml:mrow> <mml:mi mathvariant="normal">e</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> </mml:math> . The real-time predictor achieved 94% successful alarm rate with an average alarm time of 137 ms at 10% false alarm rate, demonstrating robustness against measurement uncertainties. An analysis of a typical LMiD discharge confirmed the model’s ability to identify critical parameter thresholds preceding disruptions.