Shimao Wang

The formation and development of fractures increase reservoir heterogeneity and improve reservoir performance. Therefore, it is of great research value to accurately identify the development of fractures. In this paper, two- and three-dimensional models are constructed based on the finite element method and compared with the real axis integration method. The influence of different geometric parameters on the Stoneley wave amplitude is studied to assess the propagation of Stoneley waves in the fracture zone in the well. The results show a significant positive correlation between the width and number of fractures and the attenuation coefficient of Stoneley waves. The fracture angle has a negative correlation with the attenuation coefficient and lesser impact on Stoneley waves. In addition, Stoneley waves are less sensitive to changes in fracture location, while the sensitivity to fracture spacing is significant in the range of 50 cm to 75 cm. The main propagation depth of Stoneley waves occurs 20 cm from the wall of the well. Quantitative analyses of the fracture width, number, location, spacing, depth, and angle are conducted to determine the influence of the fracture parameters on the Stoneley wave attenuation coefficient, clarify Stoneley wave propagation in wells, and provide a theoretical basis for the accurate evaluation of fractures.

Capillary pressure curves are usually obtained through mercury injection experiments, which are mainly used to characterize pore structures. However, mercury injection experiments have many limitations, such as operation danger, a long experiment period, and great damage to the sample. Therefore, researchers have tried to predict capillary pressure data based on NMR data, but NMR data are expensive and unstable to obtain. This study aims to accurately predict capillary pressure curves. Based on rock particle size data, various machine learning methods, such as traditional machine learning and artificial neural networks, are used to build prediction models and predict different types of capillary pressure curves, aiming at studying the best prediction algorithm. In addition, through adjusting the amount of particle size characteristic data, the best amount of particle size characteristic data is explored. The results show that three correlation coefficients of the four optimal algorithms can reach more than 0.92, and the best performance is obtained using the Levenberg–Marquardt method. The prediction performance of this algorithm is excellent, with the three correlation coefficients being all higher than 0.96 and the root mean square error being only 5.866. When partial particle size characteristics are selected, the training performance is gradually improved with an increase in the amount of feature data, but it is far less than the performance of using all the features. When the interpolation increases the particle size characteristics, the best performance is achieved when the feature data volume is 50 groups and the root mean square error is the smallest, but the Kendall correlation coefficient decreases. This study provides a new way to obtain capillary pressure data accurately.