Intelligent Location of Microseismic Events Based on a Fully Convolutional Neural Network (FCNN)

As a 3D real-time monitoring method, microseismic (MS) monitoring technique has been widely used in various underground engineering applications. However, in such applications, it is still challenging to acquire precise and efficient MS locations. Here, we examined the applicability and accuracy of a fully convolutional neural network for source localization, where the modified loss function was utilized. The Shuangjiangkou underground powerhouse in southwestern China served as the engineering background. The dataset was made of the MS events that occurred near the main powerhouse from September 2018 to December 2019. A fully convolutional neural network, named MS-location Net, was then built. The original waveform data were directly used as the input of the neural network, while 3D Gaussian distribution functions of the monitoring area were used as the output of the neural network. The epicenter error, focal depth error and absolute error were applied as indicators to evaluate the model. The results show that all the three indicators, namely the epicenter error, focal depth error and absolute error, were less than 5 m for all the MS events in the test set. The average time for locating an MS event was 0.01435 s using a usual computer configuration, which greatly improves the positioning efficiency. The proposed location method in this paper overcomes the shortcomings of the traditional localization methods, e.g., the inaccuracy of velocity model and arrival picking. Highlights A fully convolutional neural network, named MS-location Net, was built for microseismic source localization. The applicability and efficiency of the proposed location method were validated via a case study. The proposed location method overcomes the shortcomings of the traditional localization methods, e.g., the inaccuracy of velocity model and arrival picking.