Research On Noise Elimination And Qualitative Analysis Of Marine Controlled-Source Electromagnetic Data

As a new geophysical exploration method based on the conductivity difference between oil and gas reservoir and surrounding rock,Marine Controlled Source Electromagnetic Method(MCSEM)can help judge whether there is oil and gas reservoir with high resistance in seabed stratum,and it has been widely adopted in the evaluation of seabed oil and gas reservoir.However,in MCSEM exploration,with the decrease of the signal,the effective electromagnetic response at the middle and long deviation distance will become very weak,which is very vulnerable to noise interference,resulting in the decreased accuracy of manual qualitative analysis.In recent years,deep learning technology has gradually become a research hot spot in the field of geophysical exploration.Therefore,the research on MCSEM data and qualitative analysis technology based on deep learning is of great research significance and application value for improving the quality of marine controllable source electromagnetic data and assisting qualitative analysis.This thesis is divided into three parts:1)Based on the MCSEM exploration theory and analysis of the MCSEM signal,and combining the one-dimensional forward program,a variety of different seabed formation models is constructed by changing the setting of exploration simulation parameters in this paper.In addition,the theoretical MCSEM signal is calculated and the theoretical dataset is established,and different degrees of simulated random noise are superimposed to establish the synthetic dataset.2)The depth convolution autoencoder noise elimination model is constructed based on the autoencoder,and according to the signal characteristics,research model optimization methods such as increasing network depth,batch normalization and attention mechanism to enhance the feature extraction ability of the network,using the theoretical dataset and synthetic dataset for model training and testing,use the signalto-noise ratio and mean square errors to verify the validity of the model and apply the model to the measured data.According to the results,the signal-to-noise ratio of the eliminated noise data increased from 19.45 d B to 38.93 d B,with the mean square error being reduced from 0.0205 to 2.3835E-04,and the measured data can still retain the signal characteristics relatively completely,and realize the noise elimination of MCSEM data.3)A qualitative analysis method of MCSEM data based on Long Short-Term Memory(LSTM)is designed,used the single-layer,double-layer and bidirectional LSTM networks structures to qualitative analysis by the theoretical dataset and the synthetic dataset respectively to determine whether there is a high-resistance layer,and calculated the accuracy of qualitative analysis and model training time to evaluate the effect.According to the results,the accurate rate of qualitative analysis of the three models in the theoretical dataset can reach 100%,and the effect of bidirectional LSTM in the synthetic dataset is better,with the accuracy rate of 80.58%.In brief,the marine controllable source electromagnetic data noise elimination and qualitative analysis technology proposed in this paper has a good capability in elimination and qualitative analysis,which is of great research significance for improving the quality of MCSEM data and assisting qualitative analysis.