Abstract: The shale oil in the Fengcheng Formation of the Mabei Area has emerged as a key successor field for reserve and production growth in the Junggar Basin. The matrix physical properties of the Fengcheng Formation reservoirs are relatively poor, with fractures serving as the primary controlling factor for high production. Additionally, differentiated supporting fracturing technologies are required for reservoirs with varying degrees of fracture development. Traditional well-seismic joint fracture modeling methods exhibit low accuracy and struggle to quantitatively characterize multi-scale, azimuthally distributed fractures. This paper innovates methods in the following aspects: 1) azimuthal fracture modeling based on well data and seismic information is conducted to enhance the pertinence of the research; 2) tectonic evolution analysis and multi-constraint stress field inversion are performed to simulate the distribution of fracture development in different geological stages driven by paleotectonic stress field; 3) through the method of deep learning, the fracture information from well logging, seismic data, and stress fields is effectively integrated into a unified framework, enabling precise characterization of fractures of different scales and periods. The application of these methods raises the accuracy of natural fracture prediction to more than 80%, providing technical support for the fracturing design of 6 wells in the M Area. The daily production capacity of individual wells has increased by more than 22%, contributing to cost reduction and efficiency enhancement in the oilfield.

Key words: shale oil, fracture modeling, seismic attribute, paleostress field, neural network fusion, fracture simulation