基于LightGBM的EAST杂质破裂预警  

Prediction of EAST Impurity Disruption Using LightGBM

作　　者：孙召宏 胡文慧 袁旗平 高彬富 丁锐 曾龙 肖炳甲[1] SUN Zhao-Hong;HU Wen-Hui;YUAN Qi-Ping;GAO Bin-Fu;DING Rui;ZENG Long;XIAO Bing-Jia(Institute of Plasma Physics,Hefei Institutes of Physical Science,Chinese Academy of Sciences,Hefei 230031,China;University of Science and Technology of China,Hefei 230026,China)

机构地区：[1]中国科学院合肥物质科学研究院等离子体物理研究所,合肥230031 [2]中国科学技术大学,合肥230026

出　　处：《计算机系统应用》2023年第1期50-60,共11页Computer Systems & Applications

基　　金：国家磁约束核聚变能发展研究专项(2018YFE0302100);国家自然科学基金(12005264,12075285,U1867222,12022511)。

摘　　要：对全超导托卡马克核聚变实验装置东方超环(EAST)运行放电期间发生的杂质破裂进行预测对未来的聚变装置的长脉冲稳态放电有重要意义.根据杂质破裂的物理特性筛选出的2018年的334炮杂质破裂炮数据以及2021年的1628炮非破裂炮作为训练炮,再由等离子体平衡、密度、电流以及辐射等8种诊断信号组成的训练样本以LightGBM算法训练出杂质破裂预测模型.实验结果表明LightGBM算法模型可以对杂质破裂进行准确预测(成功预测率96.29%),非破裂炮的误判率6.87%.研究结果证明利用LightGBM进行EAST等离子体杂质破裂预警是可行的方案.The prediction of impurity disruption during the discharge period of experimental advanced superconducting tokamak(EAST)is of great significance for the long-pulse steady-state discharge of future EAST.According to the physical characteristics of impurity disruption,the data of 334 impurity disruptive discharges in 2018 and 1628 nondisruptive discharges in 2021 are selected as training discharges.Then,the training samples composed of eight diagnostic signals,including plasma equilibrium,density,current,and radiation signals,are used to train the impurity disruption prediction model by LightGBM.The test results reveal that the LightGBM model can accurately predict the impurity disruption,with a success rate of 96.29%,while for non-disruptive discharges,the false positive rate is 6.87%.The research results indicate that it is feasible to use LightGBM to predict plasma impurity disruption of EAST.

关 键 词：全超导托卡马克 EAST 杂质破裂 破裂预警 LightGBM 

分 类 号：TL631.24[核科学技术—核技术及应用]