Abstract: Abstract: [Objective]Automatic pickup of effective events is an important part of microseismic monitoring, and the accuracy of pickup directly affects the accuracy and reliability of subsequent seismic source localization and seismic source mechanism inversion. [Methods] In this paper, a 10-layer U-Net neural network model framework is constructed, the original microseismic data from 3D finite-difference simulation and the raw microseismic data from the measured gas storage reservoirs are made into labeled images, which are cut into 128*128 sized slices and input into the U-Net neural network for learning, and then the output of predicted slices is outputted and merged, and then the predicted images are binarized, and the microseismic effective events are extracted in the end of the P-wave first arrivals. This makes the edge segmentation of background noise and effective signal image more fine, and improves the efficiency and accuracy of automatic picking up of effective microseismic events.[Results]Quantitatively analyze and compare the pickup rate, wrong pickup rate and pickup error of U-Net method and STA/LTA method, the test results show that the pickup effect of U-Net is better than that of STA/LTA method, and U-Net also has a strong anti-jamming ability and generalization ability; Evaluate the effect of different label widths on the first-to-pickup results, the results show that the label pickup effect based on the event's primary cycle is The results show that the label pickup effect based on the main cycle of the event is the best.[Conclusion] The U-Net neural network first-to-automatic pickup algorithm established in this paper is an important part of the highly efficient and high-precision reservoir integrity microseismic intelligent monitoring system, which is of great significance to improve the level of microseismic monitoring technology in China.