RBM-based seismic prestack signal clustering method

Abstract

The invention discloses an RBM-based seismic prestack signal clustering method. Aiming at the characteristic of high dimensionality of seismic prestack signals, a dimension reduction method based on adepth limited Boltzmann machine network is introduced, a depth limited Boltzmann machine has relatively good abstraction feature extraction ability, and the output after dimension reduction can wellexpress an original signal. Aiming at the characteristic that the seismic signal is often noisy and hard to remove well, a fuzzy self-organizing neural network clustering algorithm is introduced, so that the seismic phase classification is expressed by membership degree; the information provided by the fuzzy classification result is more abundant and reasonable, so that the clustering result is more stable, and the classification result can be further explored. The method of the invention has lower computational complexity and can be popularized in large-scale seismic prestack data.

Description

technical field [0001] The invention belongs to the technical field of earthquakes, and relates to seismic signal classification technology, in particular to a seismic pre-stack signal clustering method. Background technique [0002] The generation of seismic phase maps by seismic signal classification technology to determine underground reservoirs plays an important role in guiding oil exploration. The existing seismic signal classification technology is mainly aimed at post-stack signals. Post-stack signals are the horizontal summation of pre-stack signals, resulting in a lot of seismic information lost. Compared with the post-stack signal, the notable feature of the pre-stack signal is that the dimension becomes higher. Directly using the existing method will bring the disaster of dimensionality, and the classification algorithm is not suitable. At the same time, the amount of data becomes large, and algorithms with high computational complexity cannot be used. After ob...

AI Technical Summary

Problems solved by technology

[0005] In clustering algorithms, it is often necessary to combine feature dimensionality reduction algorithms, because high-dimensional data will make clustering algorithms unsuitable and difficult to interpret.

Images