Prestack time-varying wavelet extraction method based on improved gated recurrent units network
Accurate extraction of seismic wavelets is the premise of subsequent inversion and imaging.In this paper,an improved Gated Recurrent Units(GRU)network for prestack time-varying seismic wavelet extraction is proposed to solve the problems that traditional time-varying wavelet extraction methods need to conform to various assumptions,need to extract amplitude and phase separately.First,the nonstationary convolution model is used to build the training data set based on the non-stationary nature and time-series properties of prestack seismic data.Then a neural network model was built to expand the extracted time-series features,including a multi-layer GRU and a fully connected neural network.Later the training data set is used to train the network model so that the network can extract time-varying wavelets.We define the specified loss function to measure the error during the backpropagation of training.Finally,achieve the accurate extraction of prestack time-varying wavelet.The synthetic data simulation experiments and the comparative wavelet extraction experiments verify that the proposed time-varying wavelet extraction method has improved the accuracy compared to the conventional methods.Using the actual prestack seismic data in western China,we showed that the proposed method improved the resolution of the seismic profiles in the target area.
NETL
NSTL
万方数据
