Shear wave prediction method of tuffaceous reservoir based on LSTM cyclic neural network

Abstract

Tuffaceous reservoir has various mineral components and complex pore structure. The traditional shear wave prediction method based on petrophysical model is inefficient and its accuracy is difficult to guar-antee. Therefore, a method of predicting S-wave velocity of tuffaceous reservoir by long-term and short-term memory neural network (LSTM) is proposed. This method can effectively learn the sequence infor-mation of logging curve by using the long-term memory function of LSTM neural network. The sensitivity analysis of logging curve is pre-ferred, and the optimal logging curve combination sensitive to shear wave velocity is optimized through correlation analysis; Then the LSTM neural network model is trained by using the optimized logging curve combination; Finally, the prediction effect is analyzed and com-pared. The application of actual logging data shows that the indexes such as root mean square error and correlation coefficient of shear wave velocity obtained by the above method are obviously better than the traditional SVM machine learning algorithm and petrophysical model method, which verifies the stability and accuracy of this method in predicting shear wave velocity of tuffaceous reservoir with complex lithology and physical properties.