CAE Neural Network and Reverse-Time Localization Imaging Method for Picking of Downhole Fracturing Microseismic Signal

Effective surface-based monitoring of microseismic signals generated by hydraulic fracturing in drilling holds significant importance for increasing oil and gas production. Microseismic monitoring methods based on travel time analysis are widely used, but real-world observational data often contain substantial noise, which limits accurate picking of microseismic signal travel times and precise localization of seismic sources. To address this, this paper integrates the Continuous Wavelet Transform (CWT), Convolutional Autoencoder (CAE), and K-means clustering algorithm to establish a CAE neural network model for microseismic waveform picking. Building on the obtained high-precision microseismic signals, a preliminary test of microseismic localization imaging based on the reverse-time localization principle was conducted. Results from simulation experiments and initial practical applications demonstrate that this method can distinguish high- and low-frequency microseismic signals under low signal-to-noise ratio (SNR) conditions, enables simultaneous localization of multiple seismic sources, and exhibits adaptability and robustness in strong-noise environments.