Microseismic identification and first-arrival picking are the fundamental parts of microseismic monitoring. A deep learning model of convolution long short-term memory network (Conv-LSTM-Unet) was proposed to reduce the manual task of microseismic identification and first-arrival picking and overcome the problems of complex parameter design, low accuracy, and poor stability of traditional methods. This paper integrates microseismic identification and first-arrival picking tasks into a semantic segmentation task by using Conv-LSTM-Unet. In order to learn the spatio-temporal characteristics of microseismic, the model is based on the Unet network, and the Conv-LSTM model is formed by combining the convolution operation with LSTM, which replaces the convolution part of the original Unet network. Meanwhile, the microseismic signal is divided into three forms of single component, time–frequency map, and three-component signal to study the effect of microseismic input form on the identification model and first-arrival picking. The results show that when the signal input is three-component form, the model recognition and first-arrival picking effect are best. The Conv-LSTM-Unet model has outperformed other traditional models in first-arrival picking, with recognition accuracy up to 96.55% and maximum error of 4 ms.
2
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
First-arrival picking is a crucial step in seismic data processing. Because of the diverse background noises and irregular near-surface conditions, it is difcult to pick frst arrivals. In addition, existing algorithms are usually sensitive to parameter settings. Therefore, this paper proposes the frst-arrival picking through fuzzy c-means and robust locally weighted regression (FPFR) algorithm consisting of two subroutines. The pre-picking subroutine obtains initial frst arrivals through fuzzy c-means clustering and adaptive cluster-selection techniques. The smoothing subroutine handles background noises and near-ground conditions through adaptive parameter regression technique. The experiment is conducted on six feld seismic datasets and one synthetic dataset. Results show that FPFR is more accurate than three state-of-the-art methods.
3
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
Most of the microseismic signals have low signal-to-noise ratio (SNR) due to the strong background noise, which makes it difficult to locate the first arrival time. Both accuracy and stability of conventional methods are poor in this situation. To overcome this problem, here we proposed a new method based on the adaptive Morlet wavelet and principal component analysis process in wavelet coefficients matrix. The three components of microseismic signal make it possible to extract the features in wavelet coefficients domain. Then the reconstructed signal from weighted features presents an obvious first arrival. Tests on synthetic signals and real data provide a solid evidence for its feasibility in low SNR microseismic signal.
JavaScript jest wyłączony w Twojej przeglądarce internetowej. Włącz go, a następnie odśwież stronę, aby móc w pełni z niej korzystać.