密度极限破裂是托卡马克等离子体的主要破裂类型之一，而高密度运行是提高未来商用聚变堆效率的必然要求，因此为了避免密度接近 Greenwald 极限时破裂对装置造成损伤，需要及时预测密度极限破裂，并采取相关预防和减轻措施。本项目计划依托 EAST、DIII-D 和 Alcator C-Mod 三个装置，用机器学习 的方法进行跨装置密度极限破裂预警研究。本项目首先根据三个装置的历史放电数据建立密度极限破裂数据库，然后用集成学习算法--随机森林对数据库进行训练。训练的第一步是在单个装置上分别训练，研究随机森林对不同装置的破裂的预测效果，第二步是用单个装置训练的模型去测试另外两个装置，研究不同装置之间预警模型的通用性，最后对三个装置的数据混合训练，以期提取密度极限破裂的一般特征，可以扩展应用到未来的聚变装置。

Density limit disruption is one of the main types of tokamak plasma disruption. High density operation is required for improving future commercial fusion reactors efficiency. In order to avoid damage caused by density limit disruption which often happens when density approaches the Greenwald limit, it is necessary to predict and give early warnings and take actions to mitigate the disruption. This project plans to carry out cross-machine density limit disruption prediction research based on EAST, DIII-D and Alcator C-Mod tokamaks by machine learning techniques. The first part of the research is to establish the density limit disruption database from history discharges of the three devices. The second part is to use random forest which is an ensemble learning method to train the database. The training process consists of three steps. the first step is training on each device, to investigate the prediction performance on single device. The second step is to test the predictor trained from one device on the other two devices, to investigate the commonality of the prediction model between devices. The third step is to mix the database of three devices for training, to extract general characteristics of density limit disruption for the purpose of applying them to future fusion devices.