论文部分内容阅读
地震记录初至的拾取可作为一个模式识别过程而引入神经网络理论进行分析。本文将一个三层感知器成功地用于地震记录初至拾取。神经网络训练采用误差反向传播算法，其中学习率在学习过程中随着输出节点的误差自动调节，隐层节点在学习过程中可以自动增加，从而加快了训练的速度；同时，也为克服网络陷入局部极小起到一定作用。该法采用了地震峰值、均方根振幅比、信噪比、前后峰值差等五个特征量进行选择。训练后的网络对不同探区的地震初至拾取的成功率可达97％以上。
The first arrival of seismic records can be used as a pattern recognition process to introduce neural network theory for analysis. In this paper, a three-layer sensor is successfully used for the first arrival of seismic records. Neural network training using error backpropagation algorithm, in which the learning rate in the learning process with the output node automatically adjust the error, hidden nodes in the learning process can automatically increase, so as to speed up the training speed; the same time, to overcome the network Caught in a local minimum played a role. The method uses the five peak value, the root mean square amplitude ratio, signal to noise ratio, before and after the peak difference of five features to choose. The trained network can achieve a success rate of more than 97% for the first arrival of earthquakes in different exploration areas.