Brittleness evaluation method of tight sandstone based on Adam-neural network: A case study of a block in Gaobei slope, Nanpu Sag

The brittleness evaluation of tight sandstone is of great importance in sweet point prediction and fracturing stimulation. To clarify the mechanics mechanism of brittleness and better the accuracy of the brittle mineral analysis, we propose an Adam-neural network brittleness evaluation method, which takes the mechanical property, mineral component and rock maturity into account. Firstly, we conducted triaxial mechanical experiments on 27 samples from the northern Nanpu Sag, and analyzed the stress-strain curve so as to obtain the brittleness index based on Rickman method. Secondly, according to the X-ray diffraction, we obtained the content of clay and non-clay, which respectively reflect the rock maturity and brittle component. Then we used an advanced Adam algorithm to form a neural network evaluation model, setting the Rickman brittleness index as objection function and mineral content as training parameters. Finally, we validated the model accuracy with the logging curve. The result shows that the brittle minerals of the region are mainly quartz and feldspar. The rock shows medium brittleness but strong anisotropy. This result is consistent with the brittleness index calculated by logging data. With all the mechanical, geological and mineralogical factors combined, this Adam-neural network model can help obtain more accurate brittleness index in a broader area, which provides a good basis for fracturing parameter optimization and target layer selection.