| With the rapid development of science and technology around the world,oil and gas resources have an important strategic position in industry.China has repeatedly stressed the importance of ensuring oil and gas security on many occasions.The purpose of seismic exploration is to try to detect the underground oil and gas resources,and determine the safe and efficient oil and gas exploitation route through comprehensive analysis of the collected data,so as to exploit the underground oil and gas resources.As an important resource storage area in China,the Tarim Basin has huge potential for exploitation.At present,there are still a large number of oil and gas resources that have not been exploited.At the same time,due to the unique geological and climate condition of the Tarim Basin,there are multiple types of noises in the collected data,and the noise and signal are overlapped greatly in the time and frequency domains,which puts forward higher requirements on noise suppression methods.After years of research and development,there are many methods for seismic noise suppression.These methods can be divided into traditional methods and neural network-based methods.Among them,the traditional method needs the prior characteristics of the data,and the selection of some parameters depends on manual setting.In contrast,the method based on neural network can learn a lot of data,and the network can adjust its parameters during the training process,so it has a good application prospect.Due to the continuous improvement of computer performance,neural network also shows a good application prospect.Neural network can improve processing performance and reduce operation speed and labor cost.This paper proposes a neural network algorithm based on multiple attention mechanisms for desert seismic noise suppression.Firstly,the algorithm uses mean shift to remove the influence of extreme values in noisy data on the overall processing effect.After that,the network adopts two attention mechanisms,namely,Enhanced Attention Module and Supervised Attention Module.The enhanced attention module focuses on learning the correlation between different channels to ensure the correlation of desert signals.The supervised attention module relearns the input noisy signals so as to emphasizes the important features again.In the enhanced attention module,it first expands the receptive field of the network by using four dilated convolutions with different dilation rates,so it can learn as much as possible from the input data.At the same time,it uses the sigmoid activation function activates the feature.The innovation of the algorithm lies in the combination of two attention mechanisms,which can more effectively extract effective features from noisy desert seismic data,and suppress noise as much as possible while preserving signal integrity.Next,this paper first verifies the noise suppression performance of the network training model by using analog signal validation,and compares it with classical denoising methods.Through the comparison of denoising pictures,F-K spectrum analysis,single channel analysis,etc.,this method shows sound noise suppression ability.While suppressing noise,it can better retain the peak and valley values of the signal,thus preserving the weak signal in the noisy data.In addition,this method also tests different speed models,proving the universality of the model.After that,the network processes the actual desert seismic data.This method can recover some weak signals in the noisy data,and also ensure the continuity of the signal axis,which once again proves that this method has good application prospects.In addition,this method also uses ablation experiments to prove the effectiveness of the combination of attention mechanisms proposed in this paper.Through the numerical analysis of signal to noise ratio and single-channel comparison,it can be seen that the design of attention mechanisms proposed in this method can ensure that the signal waveform does not change while having a good signal to noise ratio. |
PDF Full Text Request