Acoustic Logging Wave Field Simulation And Formation Velocity Inversion Based On Neural Network

Acoustic logging inversion is to get wave characteristics(wave velocity,attenuation and dispersion of each wave group)from downhole measurement waveform data signal processing,and then invert formation parameters(porosity,permeability and saturation)based on the knowledge of rock physics.But in the slow formation,there is no S-wave group in the sound pressure curve received by the receiver.Because neural network can fit nonlinear function or make logical decision only by training data.Therefore,this thesis introduces the neural network,in the case of slow formation can not receive shear wave group,using neural network to inversion formation velocity.Due to the lack of actual downhole measurement waveform data,this thesis first establishes the acoustic logging model with monopole source and dipole source as sound source from the wave equation of elastodynamics,and synthesizes the acoustic pressure waveform curve according to the formation P-wave,S-wave velocity and density in oilfield exploration.The synthesized sound pressure waveform is integrated into neural network data set.The neural network is trained so that it can output P-wave and S-wave velocities of elastic formation under the condition of inputting full wave sound pressure waveform curve.Secondly,the prediction results of seven kinds of neural networks(multilayer perceptron,Alexnet,Vggnet,Inception,Resnet,Lenet and recurrent neural networks)in inversion of formation velocity on monopole and dipole source data sets are compared.The results show that the three neural networks with the best prediction effect are multilayer perceptron,Alexnet and recurrent neural network;The overall performance of neural network on monopole data set is better than that on dipole data set.Finally,a more simple and efficient neural network structure is designed according to the principle of Alexnet,and the monopole data set is input for training.The mean square error is 219.741,and the coefficient of determination between the predicted data and the real data is 0.999841.The neural network method is compared with STC(Slowness-Time Coherence method)commonly used in acoustic logging.The results show that the average relative error of S-wave velocity predicted by neural network method is 0.00689,and that of P-wave velocity predicted by neural network method is0.00406,which is lower than that of STC method;More importantly,the neural network method can invert the S-wave velocity of slow formation,but the STC method can't.