Abstract: Based on the deep learning method, this paper introduced an ELM real-time recognition algorithm on HL-2A tokamak. The algorithm used data from 5200 shots (about 241,900 data slices) for learning and a 22-layer convolutional neural network was obtained. The algorithm has recognized all historical data of HL-2A since it achieved stable ELMy H-mode discharge in 2009. A total of 1665 shots of H-mode have been recognized, of which 35 shots were misidentified, with the false positive rate (FPR) of 2.10%. In the actual 1634 shots of H-mode, the system missed to recognize 4 of them, with the false negative rate (FNR) of 0.24%. The FPR and FNR fulfill the precision requirements of real-time ELM recognition. In the simulated real-time environment, the algorithm’s average calculation time of a single slice is 0.46 milliseconds, which satisfy the calculation speed requirements of the real-time ELM recognition.

Key words: H-mode, Neural network, Deep learning, HL-2A tokamak.

摘要： 基于深度学习的方法，在HL-2A装置上开发出了一套边缘局域模(ELM)实时识别算法。算法使用5200次放电数据(约24.19万数据切片)进行学习，得到一个深度为22层的卷积神经网络。为衡量算法的识别能力，识别了HL-2A装置自2009年实现稳定ELMy H模放电以来所有历史数据(约26000次放电数据)，共识别出1665次H模放电，其中误识别35次，误报率为2.10%。在实际的1634次H模放电中，漏识别4次，漏识别率为0.24%。该误报率和漏报率可以满足ELM实时识别的精度要求。识别算法在实时控制环境下，对单个时间点的平均计算时间为0.46ms，可以满足实时控制的计算速度要求。

关键词: H模, 神经网络, 深度学习, HL-2A装置