Research On Denoising Method Of Microseismic Signal Based On Convolutional Neural Network

The microseismic monitoring technology monitors the microseismic signals generated by underground rock fractures,identifies and inverts the microseismic signals,understands the trend of rock fractures,and helps improve the permeability of low-permeability reservoirs and increase the production of unconventional oil and gas.However,the artificially induced microseismic signal is weak and easily interfered by the surrounding noise,which brings certain difficulties to the identification and application of the microseismic signal.Therefore,suppressing the noise in the microseismic signal is the key to the microseismic monitoring technology,which is of great significance to the acquisition of energy and the development of the country.In recent years,with the widespread application of artificial intelligence technology,convolutional neural networks have also been developed rapidly,and their applications in the field of noise reduction have received extensive attention.According to the microseismic signal and its existing noise characteristics,this paper has carried out in-depth research on the noise reduction method of microseismic data based on convolutional neural network.The specific research content mainly includes the following three aspects.First,a convolutional neural network microseismic noise reduction method based on transfer learning is proposed.The model is pre-trained with data in a certain field.The pre-trained model can quickly extract the characteristics of the microseismic data,speed up the model training,and solve the problem of insufficient microseismic data.Then,the model is trained through the microseismic data set,and the convolutional neural network model is fine-tuned.Further improve the noise reduction performance of the convolutional neural network model.By comparing with the wavelet threshold noise reduction algorithm,the convolutional neural network model proposed in this paper can reduce the noise in the microseismic signal,restore the amplitude of the microseismic signal,and process the edge signal well.Second,a fast convolutional neural network microseismic noise reduction method is proposed.The convolutional neural network achieves performance improvement by stacking convolutional layers,but it also brings an increase in the amount of calculation,which leads to a decrease in calculation efficiency and cannot feed back the noise reduction results of microseismic signals in real time.In view of this,we use deep separable convolution instead of traditional convolution to reduce the parameter amount of the model;reduce the dimensionality of the sample image of the convolutional neural network by sampling the input data to reduce the dimensionality of the sample image of the convolutional neural network,thereby improving the processing speed of the model.Third,it verifies the ability of the proposed model to deal with complex noise.By mixing different levels of Gaussian noise and salt and pepper noise,a more complex new type of noise is obtained,and then the noise reduction effect of the proposed model on this type of complex noise is investigated.Simulation experiments show that the fast convolutional neural network model not only has a certain degree of reduction in both the amount of parameters and the amount of calculation,but also the effect of noise reduction is relatively ideal.This paper has conducted a preliminary exploration on the noise reduction of microseismic,and the research results reveal that the research scheme of using convolutional neural network to implement the noise reduction of microseismic is feasible.This method has certain reference significance in enriching the theory of microseismic data processing and expanding the engineering practice of microseismic data noise reduction based on artificial intelligence.