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

Microseismic monitoring technology is the main means of unconventional oil and gas development,and its monitoring results provide a very important guarantee for improving shale gas exploration technology and unconventional oil and gas recovery.However,the actual microseismic signals often contain a lot of noise.The existence of noise reduces the accuracy and reliability of subsequent microseismic data interpretation.Therefore,one of the key steps of microseismic monitoring is to reduce the noise in microseismic signals.In recent years,with the rapid development of deep learning,recurrent neural networks have been widely used.By studying the characteristics of microseismic signal and recurrent neural network,it is considered that it is feasible to use recurrent neural network to reduce the noise of microseismic data.Therefore,the method of reducing noise of microseismic data based on recurrent neural network is studied in depth.The specific research content mainly includes the following two aspects.First,a deep bidirectional gated recurrent unit recurrent neural network based microseismic data denoising method is proposed.Firstly,the deep bidirectional gated recurrent unit recurrent neural network model is constructed by using gated recurrent unit and three-layer bidirectional network structure.Secondly,ricker wavelet forward modeling was used to synthesize the pure microseismic data set.Different levels of Gaussian noise and different types of noise were added to the pure microseismic sample set to construct the noise-containing microseismic data set,and the network model was trained by the constructed data set.Then,the denoising results of the deep bidirectional gated recurrent unit recurrent neural network model are compared with the denoising results of gaussian filtering,BP neural network,long and short memory neural network and traditional recurrent network.The experimental results show that the deep bidirectional gated recurrent unit recurrent neural network model can effectively reduce the noise in the microseismic signal and restore the microseismic signal.Second,a method of microseismic noise reduction based on one-dimensional convolutional recurrent neural network is proposed.Firstly,a one-dimensional convolutional recurrent neural network model is constructed.Secondly,the mixed noises of different levels of Gaussian noise and different types of noise are added in the construction of noisy microseismic data set to verify the ability of the model to deal with multiple types of noise.Then train the network model.The results of one-dimensional convolutional recurrent neural network denoising are compared with those of wavelet denoising,gated recurrent unit recurrent neural network and deep bidirectional gated recurrent unit recurrent neural network.Finally,the number of model parameters and reasoning time of one-dimensional convolutional recurrent neural network and deep bidirectional gated recurrent unit recurrent neural network are compared.The experimental results show that the one-dimensional convolutional recurrent neural network can reduce the number of parameters effectively and improve the computational efficiency of the model.The above two methods are used to study the noise reduction of microseismic data by using recurrent neural network.A large number of research results show that the method of using recurrent neural network to achieve the noise reduction of microseismic data is feasible.In addition,pure microseismic signal is the basis of microseismic monitoring technology,and the method of using recurrent neural network to denoise microseismic data provides a new idea for denoising microseismic data.It is of great significance for interpretation and processing of microseismic data and development of unconventional oil and gas reservoirs.