Factor Analysis of Affecting the Accuracy for Intelligent Picking of Seismic First Arrivals with Deep Learning Model

Picking up the Seismic first-arrival wave has always been a basic and important thing in seismic data processing. However, in the case of large amount of data and low signal-to-noise ratio, picking the first-arrival wave entirely by hand has the following problems: the efficiency is low, and it is difficult to meet the requirements of timeliness in production. Based on the U-Net convolutional neural network, a supervised learning model and a transfer learning model for first-arrival wave picking were established, and a supervised learning and transfer learning sample library for first-arrival wave picking suitable for the experimental work area was constructed. Supervised learning and transfer learning model training were carried out through hyperparameter tuning, and the model prediction results were evaluated according to the error of statistical prediction result of the model test set data. It has been proved by oilfield application that the supervised learning model, transfer learning model and intelligent picking method can meet the actual needs of the oilfield in terms of the prediction accuracy and picking efficiency of the first-arrival wave picking of the 3D seismic data in the work area, and have the value of widespread application in oil and gas fields.