Shallow Bore-hole Microseismic Data Processing Based on Machine Learning

In recent years, many risks arisen during the exploitation of geothermal resources, gas storage and CCUS, which can be detected effectively by long-term microseismic monitoring. However, most of microseismic processing methods was developed for short-term monitoring during hydro-fracturing, and has limitations of high acquisition costs and low automation when applied to the long-term monitoring. An automated processing flow based on machine-learning is established by this paper, meeting the application of long-term microseismic monitoring. This flow is applied to microseismic data of hot dry rock fracturing acquired by shallow borehole receivers, and benchmarked with the traditional processing method, verifying the reliability of machine-learning methods in long-term microseismic monitoring. Our result shows a high data quality of shallow bore-hole receivers. Compared with traditional methods, the denoising and phase picking methods based on neural network achieve much higher S/N ratio and PS picking accuracy. The wrong picks are excluded by automatic quality control method based on Gaussian Mixture Model effectively. And the relocation result is similar to the traditional method with manually quality control. Combined with shallow bore-hole receivers, the automatic processing flow in this study has a good potential in the long-term microseismic data processing.