基于循环神经网络的微地震有效信号自动识别

Automatic recognition of effective signals for microseismic based on recurrent neural network

王博爱,杨瑞召,李德伟,张都,郭嘉梁

WANG Boai, YANG Ruizhao, LI Dewei, ZHANG Du, GUO Jialiang

中国矿业大学（北京）地球科学与测绘工程学院，北京 100083

College of Geosciences and Surveying Engineering, China University of Mining and Technology (Beijing), Beijing 100083, China

微地震地面实时监测时，代表压裂主要信息的有效信号能量弱，常淹没在噪声中，这将直接影响微地震事件的识别和震源定位的效果，因此，采用准确、快速的方法从微地震数据中自动识别有效信号尤为关键。文中采用长短时记忆网络（LSTM）的循环神经网络模型，通过综合微地震有效信号的能量特征、频谱特征、统计特征等属性来进行模型训练，并加入ReLU激活函数和L2正则化的损失函数来进行模型参数调优。将文中训练好的模型应用到实际微地震数据的结果表明，循环神经网络模型识别效果良好，充分压制了噪声，优化了数据成像。

 In real-time monitoring of microseismic ground, the effective signal energy representing the main information of fracturing is weak and often submerged in noise, which will directly affect the identification of microseismic events and the effect of focal location. Therefore, it is particularly critical to automaticly identify the valid signal from microseismic data by accurate and fast methods. In this paper, the recurrent neural network model with long and short-term memory network(LSTM) is used to carry out model training by integrating multi attributes such as energy characteristics, spectrum characteristics, and statistic characteristics of microseismic effective signal. And the ReLU activation function and L2 regularized loss function are used to optimize the model parameters. After applying the trained model to the actual microseismic data, the good recognition effect of recurrent neural network  model is obtained, the noise is fully suppressed, and the data imaging is optimized.

10.6056/dkyqt202105014