In this study, a long short-term memory (LSTM) model is trained on a large disruption warning database to predict the disruption on EAST tokomak. To compare the performance of the proposed model with the previously reported full convolutional neural network (CNN) (Guo et al 2020 Plasma Phys. Control. Fusion 63 025008), the same data set and diagnostic signals are used. Based on the test set, the area under the receiver operating characteristic curve, i.e. the AUC value of the LSTM model is obtained as 0.87, and the true positive rate (TPR) is ~87.5%, while the false positive rate (FPR) is ~15.1%. Since the LSTM model is more sensitive to radiation fluctuations than CNN, the prediction performance of LSTM model is inferior to that of CNN model (for CNN, AUC ~0.92, TPR ~87.5%, FPR ~6.1%). However, the advance warning time of LSTM model is 14 ms earlier than that of CNN. To reduce the FPR and improve the performance of the model, more fast bolometer channels are added as the input signals of the LSTM model, including the radiation from the upper and lower edges and the plasma core. Consequently, for the same test set, the AUC value increases to 0.89, and the FPR decreases to ~9.4%, but the TPR also decreases to ~83.9%. In addition, the sensitivity of the model to radiation fluctuations caused by impurity behavior decreases significantly, and the warning time becomes 8.7 ms earlier as compared to that of the original model. Overall, it is proved that deep learning algorithms exhibit immense application potential in the disruption prediction of long-pulse fusion devices.

In this study, a full convolutional neural network is trained on a large database of experimental EAST data to classify disruptive discharges and distinguish them from non-disruptive discharges. The database contains 14 diagnostic parameters from the ∼104 discharges (disruptive and non-disruptive). The test set contains 417 disruptive discharges and 999 non-disruptive discharges, which are used to evaluate the performance of the model. The results reveal that the true positive (TP) rate is ∼ 0.827, while the false positive (FP) rate is ∼0.067. This indicates that 72 disruptive discharges and 67 non-disruptive discharges are misclassified in the test set. The FPs are investigated in detail and are found to emerge due to some subtle disturbances in the signals, which lead to misjudgment of the model. Therefore, hundreds of non-disruptive discharges from training set, containing time slices of small disturbances, are artificially added into the training database for retraining the model. The same test set is used to assess the performance of the improved model. The TP rate of the improved model increases up to 0.875, while its FP rate decreases to 0.061. Overall, the proposed data-driven predicted model exhibits immense potential for application in long pulse fusion devices such as ITER.