基于深度信念神经网络的微震波到时拾取方法

Deep belief neural networkbased arrival picking for microseismic data

摘要: 微震信号到时的准确拾取是进行震源定位等技术的基础。传统的到时拾取方法大多依赖于人工干预，对采集数据信噪比的要求较高。为了提高低信噪比中微弱信号的拾取准确率，提出一种基于S变换和深度信念神经网络的拾取方法。该网络模型训练分两步进行，首先对经S变换处理过的原始数据利用受限玻尔兹曼机进行无监督预训练，得到网络模型参数的初值；再通过误差反向传播来微调网络参数，构建最终的深度信念神经网络模型。本次利用训练好的网络对数据进行拾取，并与STA/LTA法的拾取结果进行对比，分析结果表明本文方法有更高的抗噪性。

Abstract: In order to improve the picking accuracy of the weak signal in low signal to noise ratio，a pickup method based on STransform and deep belief neural network（DBN）was proposedThe network model training was divided into two stepsFirst，the original data processed by STransformation was subjected to unsupervised pretraining by using a restricted Boltzmann machine（RBM），and the initial values of the network model parameters were obtainedNetwork parameters adjusted by error backpropagation were used to build the final DBN modelThen，the trained network was used to pick up the dataCompared with pickup results of the STA/LTA method，the method we used has higher antinoise performance