Fast and accurate magnitude prediction is the key to the success of earthquake early warning (EEW). However, it is difficult to significantly improve the performance of magnitude prediction by empirically defined characteristic parameters. In this study, we have proposed a new approach (EEWNet) based on deep learning to predict magnitude for EEW. The initial few seconds of P-wave recorded by a single station without any preprocessing is used as the input to EEWNet, and the maximum displacement for the whole record is predicted and by which the magnitude is calculated. A large number of borehole underground strong motion records are used to train, validate and test the proposed EEWNet, and the predicted results are compared against those by empirical peak displacement P_d method. The comparison demonstrates that EEWNet produces better and quicker results than those by P_d, and EEWNet can predict magnitude between 4.0 and 5.9 as early as the first 0.5 s P-wave arrives. EEWNet is therefore expected to significantly enhance the accuracy and speed of magnitude estimation in practical regional EEW systems.

快速准确的震级预测是地震预警（EEW）成功的关键。然而，通过经验定义的特征参数很难显着提高幅度预测的性能。在这项研究中，我们提出了一种基于深度学习的新方法（EEWNet）来预测 EEW 的幅度。将单站记录的最初几秒的P波未经任何预处理作为EEWNet的输入，预测整条记录的最大位移并计算震级。利用大量钻孔地下强震记录对所提出的EEWNet进行训练、验证和测试，并将预测结果与经验峰值位移P _{d的预测结果进行比较}方法。比较表明，EEWNet 产生的结果比P _d的结果更好更快，并且 EEWNet 可以在第一个 0.5 s P 波到达时预测 4.0 到 5.9 之间的幅度。因此，EEWNet 有望显着提高实际区域 EEW 系统中幅度估计的准确性和速度。