Neuro-evolutionary event detection technique for downhole microseismic surveys

Recent years have seen a significant increase in borehole microseismic data acquisition programs associated with unconventional reservoir developments such as hydraulic fracturing programs for shale oil and gas. The data so acquired is used for hydraulic fracture monitoring and diagnostics and therefore, the quality of the data in terms of resolution and accuracy has a significant impact on its value to the industry. Borehole microseismic data acquired in such environments typically suffer from propagation effects due to the presence of thin interbedded shale layers as well as noise and interference effects. Moreover, acquisition geometry has significant impact on detectability across portions of the sensor array. Our work focuses on developing robust first arrival detection and pick selection workflow for both P and S waves specifically designed for such environments. We introduce a novel workflow for refinement of picks with immunity towards significant noise artifacts and applicability over data with very low signal-to-noise ratio provided some accurate picks have already been made. This workflow utilizes multi-step hybrid detection and classification routine which makes use of a neural network based autopicker for initial picking and an evolutionary algorithm for pick refinement. We highlight the results from an actual field case study including multiple examples demonstrating immunity towards noise and compare the effectiveness of the workflow with two contemporary autopicking routines without the application of the shared detection/refinement procedure. Finally, we use a windowed waveform cross-correlation based uncertainty estimation method for potential quality control purposes. While the workflow was developed to work with the neural network based autopicker, it can be used with any other traditional autopicker and provides significant improvements in pick detection across seismic gathers.