Full text loading...
This study investigates the impact of data preprocessing techniques on microseismic event detection using Distributed Acoustic Sensing (DAS) data. DAS technology uses fiber optic cables as seismic sensor arrays, crucial for monitoring subsurface activity in various industries. The research utilizes a dataset of microseismic events and background noise from a shale reservoir.
The data was processed in three categories: normalization only, median removal followed by normalization, and median removal, band-pass filtering, and normalization. A convolutional neural network was trained on this data to classify events. The results indicate that while minimal preprocessing can achieve high accuracy, more extensive preprocessing, particularly with band-pass filtering, improves generalization and reduces overfitting. The study concludes that the optimal level of preprocessing depends on the specific application and that longer training can enhance model performance, even with minimal preprocessing.