基于循环神经网络的微地震数据降噪方法

针对微地震信号中存在大量噪声干扰, 导致其识别困难的问题, 提出一种深度双向门控循环单元循环神经网络的方法, 并将其应用于微地震数据降噪中. 首先, 构建多层双向门控循环单元循环神经网络模型, 并设计该模型的网络结构及训练算法; 然后, 采用Ricker子波正演模拟微地震数据验证模型的有效性, 并将该方法与其他4种方法进行对比； 最后, 将真实的含噪声微地震数据输入到训练好的模型中, 即可得到降噪后的微地震数据. 仿真实验结果表明, 利用该方法降噪后与降噪前信号的峰值信噪比相比约提高36 dB, 且信号之间的相关系数值由0.088 6上升至0.933 5. 实际应用结果也表明, 该方法可有效降低实际微地震数据中的噪声.

Denoising Method for Microseismic Data Based on Recurrent Neural Networks

Aiming at the problems of the difficulty of identification casused by a  large number of noise interference in microseismic signals, we proposed a deep bidirectional gated recurrent unit recurrent neural network model and applied to microseismic data denoising. Firstly, we constructed a multi-layer bidirectional gated recurrent unit recurrent neural network model, and designed the network structure and training algorithm of the model. Secondly,  the validity of the model was verified by using Ricker wavelet forward modeling microseismic data, and  the proposed method was compared with the other four methods. Finally, by inputting the real microseismic data with noise into the trained model, the microseismic data without noise could be obtained. The simulation results show that the peak signal-to-noise ratio of the signal after denoising by the proposed method is about 36 dB higher than that before denoising, and the correlation coefficient value between the signals increases from 0.088 6 to 0.933 5. The practical application results also show that the proposed method can effectively reduce the noise in the actual microseismic data.