CN-122019970-A - Denoising method for ocean magnetotelluric data and related equipment

CN122019970ACN 122019970 ACN122019970 ACN 122019970ACN-122019970-A

Abstract

The invention relates to the technical field of geophysical exploration signal processing, and discloses a denoising method and related equipment for marine magnetotelluric data, wherein the denoising method comprises the steps of constructing a training data set and training a model to obtain a dual-path noise recognition model and a generated reconstruction model; the method comprises the steps of preprocessing and windowing new ocean magnetotelluric original data to be processed to obtain a plurality of data windows, carrying out parallel noise recognition on each data window to obtain a periodic noise mask and a random impulse noise mask, carrying out fusion processing to obtain a comprehensive noise mask, marking noise data points needing to be reconstructed in the data windows by the comprehensive noise mask, carrying out signal reconstruction on the data windows based on a generated reconstruction model and the comprehensive noise mask to obtain a processed clean data window, and carrying out splicing processing to obtain a complete denoised ocean magnetotelluric time sequence. The invention solves the defects of the existing MMT data processing technology in coping with long-period and multi-type mixed noise.

Inventors

ZHOU LIYE
HE XIAOYAN
ZHAO YUNSHENG
ZHAO DONGFENG
HU NAN
LI HAO
WANG LEI
LI XIAOYUE

Assignees

地球脉动(宁波)科技有限公司
宁波东方理工产业技术研究有限公司

Dates

Publication Date: 20260512
Application Date: 20260123

Claims (10)

1. A method for denoising marine magnetotelluric data, comprising: constructing a training data set and training a model to obtain a dual-path noise identification model and a generation type reconstruction model; Preprocessing and windowing new ocean magnetotelluric original data to be processed to obtain a plurality of data windows; Carrying out parallel noise recognition on each data window based on the dual-path noise recognition model to obtain a periodic noise mask and a random impulse noise mask; performing fusion processing based on the periodic noise mask and the random impulse noise mask to obtain a comprehensive noise mask, wherein the comprehensive noise mask marks noise data points needing to be reconstructed in a data window; Performing signal reconstruction on the data window based on the generated reconstruction model and the comprehensive noise mask to obtain a processed clean data window; And (3) performing splicing treatment on all clean data windows to obtain a complete denoised marine magnetotelluric time sequence.
2. The method of denoising marine magnetotelluric data according to claim 1, wherein constructing the training data set comprises obtaining a pair of noisy marine magnetotelluric time series x_noise and a corresponding clean marine magnetotelluric time series y_clean, and dividing the pair of noisy marine magnetotelluric time series x_noise and the corresponding clean marine magnetotelluric time series y_clean into fixed length windows with overlapping portions to form training sample pairs.
3. The method for denoising marine magnetotelluric data according to claim 2, wherein the dual-path noise recognition model is used for recognizing periodic noise and random impulse noise, and wherein the specific process of training the dual-path noise recognition model comprises periodic noise recognition model training and random impulse noise recognition model training: the periodic noise recognition model training comprises the steps of constructing a one-dimensional convolutional neural network, taking paired noisy marine magnetotelluric time sequences X_noise as input, and training by taking a periodic noise Mask mask_p generated by calculating an I X_noise-Y_clean I difference value and combining frequency domain analysis as a label, wherein the long periodic noise position in the periodic noise Mask mask_p is marked as 1, and the rest is 0; The random impulse noise identification model training comprises the steps of constructing a self-encoder network, taking a corresponding pure ocean magnetotelluric time sequence Y_clean as an input and training target to train, enabling the model to learn normal ocean magnetotelluric signal characteristics, and identifying random impulse noise through a reconstruction error by the self-encoder; the generating type reconstruction model is used for reconstructing signals of a noise area based on context information, and the specific process of training the generating type reconstruction model comprises the steps of constructing a sequence generation model which is a two-way long-short-term memory network or a transform self-attention mechanism deep learning model, and training by taking X_masked obtained by masking a paired noisy ocean magnetotelluric time sequence X_noise as input and a corresponding pure ocean magnetotelluric time sequence Y_clean as a label to enable the model to recover pure signals based on the context information.
4. The method for denoising marine magnetotelluric data according to claim 1, wherein the preprocessing and windowing mode is consistent with the window segmentation mode in the process of constructing the training data set, namely, the preprocessing and windowing mode is divided into fixed-length windows with overlapping parts.
5. The method for denoising marine magnetotelluric data according to claim 1, wherein each data window is input into a trained 1D-CNN model to obtain a predicted periodic noise mask m_p, each data window is input into a trained self-encoder model, and a random impulse noise mask m_i is obtained by calculating a reconstruction error and setting a threshold value.
6. A method of denoising marine magnetotelluric data according to claim 1, wherein the fusion process is a logical or operation of a periodic noise mask m_p and a random impulse noise mask m_i, resulting in a final integrated noise mask m_total that marks all noise data points within the current window that need to be reconstructed.
7. The method for denoising marine magnetotelluric data according to claim 1, wherein the reconstructing of the data window based on the generated reconstruction model and the integrated noise mask comprises masking the data window based on the integrated noise mask m_total, and inputting the processed sequence into the generated reconstruction model to obtain a clean data window filling the noise region.
8. A method of denoising marine magnetotelluric data as defined in claim 1 wherein the stitching process is a smooth weighted average of the overlapping regions of all clean data windows followed by stitching.
9. A denoising system of marine magnetotelluric data, comprising: The data model training module is used for constructing a training data set and training a model to obtain a dual-path noise identification model and a generated reconstruction model; The data preprocessing windowing module is used for preprocessing and windowing new ocean magnetotelluric original data to be processed to obtain a plurality of data windows; The parallel noise recognition module is used for carrying out parallel noise recognition on each data window based on the dual-path noise recognition model to obtain a periodic noise mask and a random impulse noise mask; The mask fusion processing module is used for carrying out fusion processing based on the periodic noise mask and the random impulse noise mask to obtain a comprehensive noise mask, and the comprehensive noise mask marks noise data points needing to be reconstructed in a data window; The signal reconstruction processing module is used for carrying out signal reconstruction on the data window based on the generated reconstruction model and the comprehensive noise mask to obtain a processed clean data window; And the data window splicing module is used for carrying out splicing treatment on all clean data windows to obtain a complete denoised marine magnetotelluric time sequence.
10. A mobile terminal comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements a method of denoising marine magnetotelluric data according to any of claims 1-8 when executing the computer program.

Description

Denoising method for ocean magnetotelluric data and related equipment Technical Field The invention relates to the technical field of geophysical exploration signal processing, in particular to a denoising method of ocean magnetotelluric data and related equipment. Background Marine Magnetotelluric (MMT) is one of the core technologies of deep sea geophysical exploration, and its data quality is directly related to the success or failure of subsea oil and gas resource exploration, natural gas hydrate investigation and deep geological structure research. However, MMT raw time-series data tends to be severely contaminated with a variety of complex noise. Noise of the land electromagnetic exploration mainly originates from high-frequency and strong electromagnetic interference (such as power frequency interference, high-voltage wires and the like) generated by human activities. However, the noise characteristics of the marine environment are quite different from them. The deep sea water has strong shielding and attenuation effects on high-frequency electromagnetic waves, so that the common strong electromagnetic interference on land is greatly weakened on the sea bottom of the deep sea. Instead, more complex noise types specific to marine environments mainly include: Random impulse noise, which is generated by transient electromagnetic radiation of a past ship, submarine discharge, abrupt change of contact state of electrodes and sediments, has strong randomness and burstiness, and is shown as isolated high-amplitude peak in time domain. Long period noise, which is particularly troublesome in MMT data, is diverse in source, complex in morphology, and can extend from tens of minutes to hours. Sources include disturbance of tidal motion to the attitude of a submarine instrument, temperature drift of the instrument itself, noise generated by reading and writing magnetic discs of the instrument, and repetitive electromagnetic interference in a specific marine environment. Meanwhile, the marine electromagnetic observation has the characteristics of high sampling rate (generally more than 100 Hz) and ultra-long observation period (which can last for days to months), and the data volume generated by a single task can reach tens of GB. This mass data feature poses a serious challenge to traditional processing methods. Currently, most methods for processing MMT data noise in industry directly use or improve self-land electromagnetic processing technologies, such as far reference channel suppression, wavelet transformation, median filtering and the like. These methods have fundamental drawbacks in dealing with marine specific noise: the low-frequency information damage is that the transition band of the passband and the stopband of the traditional frequency domain filtering method (such as a band stop or a low-pass filter) is difficult to be steep when the low-frequency signal is processed aiming at long-period noise. Since many important deep geological information is also reflected in the low frequency part of the signal, the use of a filter inevitably filters out the effective signal together with noise, resulting in a permanent loss of the deep information, severely affecting the depth and accuracy of the subsequent inversion. Model hypothesis limitations conventional approaches are mostly based on ideal signal or noise prior models. For example, simple fitting and stripping methods (such as polynomial fitting) can only handle trending noise with simple morphology, and cannot handle non-sinusoidal periodic noise with complex morphology. In addition, after the noise data segment is removed, the commonly adopted linear or spline interpolation method is essentially mathematical filling, the geophysical law contained in the signal cannot be restored, and when the noise segment is longer, the interpolation result is severely distorted, and the continuity and the frequency spectrum structure of the signal are damaged. The processing efficiency is bottleneck, the traditional method depends on a large amount of manual interaction and parameter tuning, has low efficiency when facing to massive high-sampling-rate data, and is difficult to realize standardized and automatic processing flows. In summary, in the prior art system, there is a lack of an automatic new paradigm that can intelligently and efficiently process long-period and random impulse noise specific to ocean, and can recover signal physical characteristics with high fidelity after denoising, and at the same time, is adequate for mass data processing tasks. Disclosure of Invention The invention aims to provide a denoising method and related equipment for marine magnetotelluric data, which are used for solving the technical problem that the existing MMT data processing technology is insufficient in coping with long-period and multi-type mixed noise. The invention is realized by the following technical scheme: in a first aspect, the present invention provides a deno