关键词: multifaceted asymmetric radiation from the edge (MARFE) movement prediction, random forest, machine learning, Experimental Advanced Superconducting Tokamak (EAST)

Abstract: Multifaceted asymmetric radiation from the edge (MARFE) movement which can cause density limit disruption is often encountered during high density operation on many tokamaks. Therefore, identifying and predicting MARFE movement is meaningful to mitigate or avoid density limit disruption for the steady-state high-density plasma operation. A machine learning method named random forest (RF) has been used to predict the MARFE movement based on the density ramp-up experiment in the 2022's first campaign of Experimental Advanced Superconducting Tokamak (EAST). The RF model shows that besides Greenwald fraction which is the ratio of plasma density and Greenwald density limit, d$\beta_{\rm p}/$d$ t$, $H_{98}$ and d $W_{\rm mhd}/$d$t$ are relatively important parameters for MARFE-movement prediction. Applying the RF model on test discharges, the test results show that the successful alarm rate for MARFE movement causing density limit disruption reaches $\sim 85%$ with a minimum alarm time of $\sim 40 $ ms and mean alarm time of $\sim 700 $ ms. At the same time, the false alarm rate for non-disruptive and non-density-limit disruptive discharges can be kept below 5%. These results provide a reference to the prediction of MARFE movement in high density plasmas, which can help the avoidance or mitigation of density limit disruption in future fusion reactors.

Key words: multifaceted asymmetric radiation from the edge (MARFE) movement prediction, random forest, machine learning, Experimental Advanced Superconducting Tokamak (EAST)