基于深度学习模式的微震信号P波自动拾取方法研究

Research on automatic picking method of microseismic signal P wave based on deep learning mode

高精度的矿山微震P波信号到时拾取是精确定位矿山微震信号的重要前提。基于微震P波信号的时域特性和计算机视觉领域中的深度学习算法,提出适用于微震P波信号的拾取模型(DMSP模型),并构建与其相适应的损失函数。该模型首先构建自适应去噪神经网络(DAA-MINE)对矿山微震信号进行去噪,其次构建分割联合拾取神经网络(Cut-SP)对微震信号P波初值和结束点进行拾取。分别采用3835和959组矿山微震信号数据作为训练集和测试集,对提出的模型进行训练和测试,结果表明:在进行DAA-MINE模型去噪后,平均信噪比提升较高,能量保留较多;与ER算法、MER算法、WFM算法、PAT-S/K算法比较而言,Cut-SP模型拾取平均误差较低,鲁棒性较强,识别速度较快,能够满足工程需要。提出的P波信号拾取模型实现了深度学习与矿山微震监测的融合,为智能化开采中微震信号的自动拾取提供了一种新方法。

High-precision pick-up of P-wave signals from mine microseismic signals is an important prerequisite for precise location of mine microseismic signals. Based on the time-domain characteristics of microseismic P-wave signals and the deep learning algorithm in the field of computer vision,this paper propose a picking DMSP method suitable for microseismic P-wave signals and constructs a suitable loss function. This method first builds an adaptive DAA-MINE denoises the microseismic signal in the mine,and then builds a segmentation joint picking Cut-SP to pick up the initial value and end point of the microseismic signal P wave. 3 835 groups and 959 groups of mine microseismic signal data are used as training set and test set,respectively,to train and test the model proposed in this paper. The results show that:after the DAA-MINE model denoising,the average signal-to-noise ratio is improved and more energy is retained;compared with the ER algorithm,the MER algorithm,the WFM algorithm,and the PAT-S/K algorithm,the Cut-SP model The average picking error is low,the robustness is strong,and the recognition speed is faster,and it can meets the engineering needs. The pickup model constructed this time realizes the integration of deep learning neural network and mine microseismic monitoring,and provides a new method for automatically picking up data of microseismic data in intelligent mining.