Scholarlink

<p>High-confinement tokamak plasmas typically exhibit characteristics such as a small aspect ratio, large elongation, and significant triangularity configuration, with the core plasma often experiencing sawtooth mode. Experimental and simulation studies have shown that the linear stability of sawtooth mode is related to the aspect ratio, triangularity, and elongation. When numerically investigating the relationship between sawtooth mode and plasma configurations, the broad parameter space and the large number of parameter combinations require substantial computational time. Therefore, this paper builds upon previous numerical simulations of the sawtooth mode instability physical model (CLT) to construct a database containing plasma configurations and their corresponding sawtooth mode growth rates. We trained three machine learning methods—KNN Random Forest and SVR—to rapidly predict the growth rates of sawtooth mode for plasma configurations across a wide parameter domain. Experiments demonstrate that the Random Forest model achieves an R² of 99.45%, while the SVR and KNN models achieve R² values of 90.38% and 97.62%, respectively. The trained models exhibit sufficient generalization capabilities. By comprehensively comparing the predictive performance of the models, the Random Forest model, which aligns best with the simulation data, was selected for prediction. The relative error between the predicted and actual values does not exceed 2%, indicating reliable prediction performance for the growth rate of sawtooth mode.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Prediction of the Linear Growth Rate of Tokamak Sawtooth Mode Based on Machine Learning and Physical Models

  • 期刊论文

  • 发表日期：

    2026年1月7日

查看全文
  • Peijie Zhang,
  • Hongwei Ning,
  • Jinhong Yang,
  • Zhenzhen Ren,
  • Sheng Liu,
  • Jun Kuang,
  • YiQing Wang,
  • Weihua Wang

摘要

High-confinement tokamak plasmas typically exhibit characteristics such as a small aspect ratio, large elongation, and significant triangularity configuration, with the core plasma often experiencing sawtooth mode. Experimental and simulation studies have shown that the linear stability of sawtooth mode is related to the aspect ratio, triangularity, and elongation. When numerically investigating the relationship between sawtooth mode and plasma configurations, the broad parameter space and the large number of parameter combinations require substantial computational time. Therefore, this paper builds upon previous numerical simulations of the sawtooth mode instability physical model (CLT) to construct a database containing plasma configurations and their corresponding sawtooth mode growth rates. We trained three machine learning methods—KNN Random Forest and SVR—to rapidly predict the growth rates of sawtooth mode for plasma configurations across a wide parameter domain. Experiments demonstrate that the Random Forest model achieves an R² of 99.45%, while the SVR and KNN models achieve R² values of 90.38% and 97.62%, respectively. The trained models exhibit sufficient generalization capabilities. By comprehensively comparing the predictive performance of the models, the Random Forest model, which aligns best with the simulation data, was selected for prediction. The relative error between the predicted and actual values does not exceed 2%, indicating reliable prediction performance for the growth rate of sawtooth mode.