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.
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.
Tokamaks are the most promising way for nuclear fusion reactors. Disruption in tokamaks is a violent event that terminates a confined plasma and causes unacceptable damage to the device. Machine learning models have been widely used to predict incoming disruptions. However, future reactors, with much higher stored energy, cannot provide enough unmitigated disruption data at high performance to train the predictor before damaging themselves. Here we apply a deep parameter-based transfer learning method in disruption prediction. We train a model on the J-TEXT tokamak and transfer it, with only 20 discharges, to EAST, which has a large difference in size, operation regime, and configuration with respect to J-TEXT. Results demonstrate that the transfer learning method reaches a similar performance to the model trained directly with EAST using about 1900 discharge. Our results suggest that the proposed method can tackle the challenge in predicting disruptions for future tokamaks like ITER with knowledge learned from existing tokamaks.
Disruption prediction using a long short-term memory (LSTM) algorithm has been developed on EAST, due to its inherent advantages in time series data processing. In the present work, LSTM is used as the model and the AUC (area under receiver operation characteristic curve) is used as the evaluation index. When the model is trained on data from the plasma current flattop phase and tested on data from the same period multiple times, the highest AUC is 0.8646 and the training time is about 6900 s per epoch. For comparison, the last 1000 ms of the flattop phases are intercepted as short time sequences. When the model is trained on data from short time sequences and tested on data from the same period, the highest AUC is increased to 0.9379 and the training time is restricted to 36 s per epoch. When the best model trained on the short time sequences is applied to the flattop phase for testing, the AUC is up to 0.9189. The experiment results show that it is possible for LSTM to train the model on data from short time sequences and migrate the model to the entire flattop phase, with a shorter training time and higher AUC value.
Plasma disruption is a very dangerous event for future tokamaks and fusion reactors. Therefore, predicting disruption is crucial for ensuring the safety and performance of reactors. In this study, the features of two deep learning algorithms are integrated to establish a multi-scale hybrid network disruption predictor. Firstly, 43 diagnostic signals are extracted by a convolutional neural network (CNN), and the time information is learned by a long short-term memory network. The predictor is trained and tested on a database containing ∼ tens of thousands non-all-metal wall discharges. Its area under the receiver operator characteristic curve (AUC), which is a common performanc metric for deep learning algorithms, reaches 0.97, and the true positive rate is ∼ 95.3% , while the false positive rate is ∼ 8%. Since EAST was upgraded in 2020, the wall condition has been upgraded from non-all-metal to all-metal. To examine the robustness of the predictor, the EAST disruption predictor is migrated to the all-metal wall experiment for the first time. It is again tested under the all-metal wall experimental data, and its warning performance decreases significantly, with an AUC of only 0.79. To improve the robustness of the predictor against disruptions, the convolutional attention mechanism is introduced into the CNN. After training and testing with the same data set, the warning performance for the all-metal wall data is improved, with the AUC value increasing to 0.84. To further improve the robustness of the predictor, the t-SNE algorithm is employed to explore the difference of the sample distribution before and after EAST upgradation, and the transfer learning algorithm is used to reduce this difference. By applying transfer learning to a small amount of discharges from the all-metal wall data, the warning performance in the all-metal wall experiment is further improved, and the AUC value increases to 0.93.
Plasma vertical displacement control is essential for the stable operation of tokamak devices. The traditional plasma vertical displacement calculation method is not suitable for balancing speed and accuracy simultaneously, which is necessary for real-time feedback control. In this study, neural networks are used to rapidly detect vertical displacement recognition. Based on a fully connected neural network, the vertical displacement calculation model is trained and tested using magnetic data of approximately 2000 shots. To compare the effects of different inputs on vertical displacement calculation, different magnetic measurement diagnostic signals are used to train and test the model. Compared with a full magnetic measurement dataset, 39 magnetic measurement signals (38 magnetic probes and plasma current) show better accuracy with mean square error <0.0005. The model is tested using historical experimental data, and it demonstrates accurate vertical displacement calculation even in the case of a vertical displacement event. In general, neural network algorithm has great application potential in vertical displacement calculation.
The magnetic diagnostic system provides input signals for the equilibrium reconstruction, plays a key role in the plasma feedback control. Real-time monitoring and abnormal signal detection of the magnetic diagnostic system are significance for the operation of tokamak. In this paper, a general fault detection algorithm is proposed, based on the spatial auto-correlation of the probes. This algorithm firstly obtains the linear mapping between the expected value of each probe and the measured values of its adjacent probe combination through regression fitting, and then sets the fault detection threshold initially based on the fitting result of the absolute error distribution. Besides, fault probe location is realized through a voting mechanism. The algorithm can simultaneously realize the fault detection of multiple non-adjacent probes. Based on the experimental data from EAST, the method described here shows a good performance and a potential in the long-pulsed operation.
In EAST long-pulsed discharge (Hundreds of seconds), electric magnetic diagnosis (EMD) is very important, since EMD not only monitors tokamak security status but also provide accurate measurement accuracy for reconstruction of plasma boundary. To avoid current measurement drift, a fiber optic current sensor (FOCS), based on the Faraday Effect, is developed and used for poloidal and plasma current feedback control for the first time, relative current measurement accuracy is within 0.5%. To ensure plasma boundary control accuracy, a detailed set of magnetic measurement calibration methods is developed before the plasma discharge. The maximum relative error is less than 1%, the corresponding control accuracy is within 1cm. To minimize integrator drift error, a long-pulse integrator test is essential, the corresponding drift error needs to be subtracted in plasma control system. Besides, the saddle coil and Mirnov coil not only help to detect MHD issues, but are also utilized for plasma disruption prediction during the long-pulse discharge.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.