The use of real-time tomography at JET opens up new possibilities for monitoring the plasma radiation profile and for taking preventive or mitigating actions against impending disruptions. By monitoring the radiated power in different plasma regions, such as core, edge and divertor, it is possible to set up multiple alarms for the radiative phenomena that usually precede major disruptions. The approach is based on the signals provided by the bolometer diagnostic. Reconstructing the plasma radiation profile from these signals is a computationally intensive task, which is typically performed during post-pulse analysis. To reconstruct the radiation profile in real-time, we use machine learning to train a surrogate model that performs matrix multiplication over the bolometer signals. The model is trained on a large number of sample reconstructions, and is able to compute the plasma radiation profile within a few milliseconds in real-time. The implementation has been further optimized by computing the radiated power only in the regions of interest. Experimental results show that, during uncontrolled termination, there is an impurity accumulation at the plasma core, which eventually leads to a disruption. A threshold-based alarm on core radiation, among other options, is able to anticipate a significant fraction of such disruptions.

在JET上使用实时断层扫描技术为监测等离子体辐射曲线以及采取预防或缓解措施以防即将发生的干扰开辟了新的可能性。通过监视不同等离子体区域（例如，芯，边缘和偏滤器）中的辐射功率，可以针对通常在重大破坏之前的辐射现象设置多个警报。该方法基于测辐射热计诊断程序提供的信号。从这些信号重建等离子体辐射轮廓是一项计算量大的任务，通常在脉冲后分析期间执行。为了实时重建辐射轮廓，我们使用机器学习来训练替代模型，该模型对辐射热计信号执行矩阵乘法。该模型经过大量样本重建训练，并能够在几毫秒内实时计算等离子体辐射曲线。通过仅在感兴趣区域中计算辐射功率，进一步优化了实现。实验结果表明，在不受控制的终止过程中，等离子体核心处存在杂质积聚，最终导致破坏。在其他选项中，基于阈值的核心辐射警报能够预测此类中断的很大一部分。