Study and Classification of Porosity Stress Sensitivity in Shale Gas Reservoirs Based on Experiments and Optimized Support Vector Machine Algorithm for the Silurian Longmaxi Shale in the Southern Sichuan Basin, China

[Image: see text] To understand the characteristics of variation in porosity and permeability, the physical properties of the shale reservoir under different stress conditions play an important role in guiding shale gas production. With the shale of the Wufeng–Longmaxi Formation in the south of the Sichuan Basin as the research object, stress-dependent porosity and permeability test, high-pressure mercury injection, and scanning electron microscope test were performed in this study to thoroughly analyze the variation in physical properties of different shale lithofacies with effective stress. Besides, the stress sensitivity of different lithofacies reservoirs was evaluated by using parameters such as pore compressibility coefficient (PCC) and porosity sensitivity exponent (PSE), while the optimized support vector machine (SVM) algorithm was adopted to predict the coefficient of reservoir porosity sensitivity. According to the research results, the porosity and permeability of shale reservoirs decline as a negative exponential function. When the effective stress falls below 15 MPa, the damage rate of permeability/porosity increases rapidly with the rise of effective stress. By contrast, the permeability curvature of the shale reservoirs plunges with the rise of effective stress. It was discovered that a higher siliceous content results in a higher permeability curvature of shale, indicating the greater stress sensitivity of the reservoir. The ratio of matrix porosity to microfracture porosity determines the PSE, which is relatively low, and low aspect ratio pores contribute to high porosity compressibility and stress sensitivity. Young’s modulus shows a negative correlation with pore compressibility and a positive correlation with Poisson’s ratio. High clay minerals have a large number of low aspect ratio pores and a low elastic modulus, which leads to both high PCC and low PSE. Based on the principal component analysis, a multiclassification SVM model was established to predict the PSE, revealing that the accuracy of the sigmoid, radial basis function (RBF), and linear kernel function is consistently above 70%. According to error analysis, the accuracy can exceed 80% with the RBF kernel function and appropriate penalty factor. The research results serve to advance the research on the parameters related to overburden pressure, porosity, and permeability. Moreover, the optimized SVM algorithm is applied to make a classification prediction, which provides a reference for shale reservoir exploration and development both in theory and practice.