应用双向长短时记忆神经网络的微地震信号降噪方法

地面微地震监测数据噪声干扰强、信噪比低，对后续的微地震初至拾取、成像定位等产生严重影响。因此，微地震信号降噪是微地震数据预处理中的关键步骤，而常规降噪方法常依赖于算法参数的设置，不具备普遍的适用性。为此，提出了一种应用双向长短时记忆(Bi-LSTM)神经网络的微地震信号降噪方法。首先，使用合成信号和实际信号构造样本数据集，对构建的Bi-LSTM模型进行训练和测试，得到降噪效果最好的模型；然后，利用训练好的Bi-LSTM网络对不同信噪比的合成信号和川渝地区油气井的实际压裂监测微地震信号进行降噪处理。降噪后的实际微地震信号用于地震发射层析成像，并分析图像以实现地面微地震信号的震源定位。实验分析结果表明，该方法能够有效降低微地震信号中的各类噪声，提高信噪比，从而提高震源定位的精度。与传统算法相比，该方法不需要参数调整，具有良好的泛化特性。

Denoising of microseismic signal based on bidirectional long short-term memory neural network

Surface microseismic signals are greatly affected by noises and have a low signal-to-noise ratio,which has a serious impact on the subsequent work of microseismic first arrival picking and imaging positioning. Therefore, the denoising of microseismic signals is a key step in the preprocessing of microseismic data. Conventional denoising methods often depend on the settings of algorithm parameters,and thus do not have universal applicability. This paper proposes a denoising method for microseismic signals based on the bidirectional long short-term memory (Bi-LSTM) neural network. First,we use synthetic signal and actual signal to construct the sample data set. By training and testing the constructed Bi-LSTM model,we obtain the model with the best denoising effect. Then,the trained network is used to denoise the synthetic signals with different signal-to-noise ratios and the microseismic signals from the actual fracturing monitor in the Sichuan-Chongqing area. The denoised actual microseismic signals are utilized for seismic emission tomography (SET), and the source location of surface microseismic is realized through analyzing the SET images. The experimental results show that the proposed method can effectively reduce various noises in microseismic signals and improve the signal-to-noise ratio,so as to improve the accuracy of source location. Compared with the traditional algorithm,the method does not depend on the adjustment of some parameters in the algorithm and has good generalization characteristics.