基于深度学习的HL-2A装置ELM实时控制系统的研制

Development of real-time ELM control system based on deep learning in HL-2A tokamak

利用HL-2A装置已开发的基于深度学习的边缘局域模(ELM)识别算法和超声分子束注入(SMBI)等ELM缓解设备组成了一个ELM实时识别和控制系统。该系统实时采集相关的输入数据,通过神经网络计算分析,输出识别信号,当检测到存在连续ELM时,触发SMBI以缓解ELM。在HL-2A装置放电实验期间对ELM实时控制系统进行了测试,识别效果明显,在1ms控制周期中,达到了ELM的实时缓解与控制。

A real-time ELM(edge localized mode)control and mitigation system was developed in HL-2A tokamak using the deep learning-based ELM identification algorithm and ELM mitigation equipment such as supersonic molecular beam injection(SMBI).The system collects relevant input data in real time,calculates and analyzes them with the neural network,and sends out trigger signals according to the output of neural network.When continuous occurrence of ELM is detected,the system will trigger SMBI to inject molecular beams to mitigate ELM.The real-time ELM control system was tested in the HL-2A tokamak experiment.It can accurately detect the ELMy H mode in the plasma discharge by analyzing the diagnostic signal within a control period of 1ms and sends control signals when ELM occurs to trigger the ELM mitigation system,and the identification effect is obvious.The system can be applied to the real-time ELM mitigation and control in HL-2A plasma.