Abstract: The ultra-deep formations in the Kuqa depression are rich in oil and gas resources, with strong reservoir heterogeneity and obvious well-to-well productivity difference. The present-day in-situ stress plays an important role in wellbore trajectory design, horizontal well operation, and fracturing design, but the current research methods are disadvantageous in many aspects. On the basis of the measured present-day in-situ stress, the relationships between in-situ stress and logging parameters were constructed, and the prediction of the present-day in-situ stress in the ultra-deep tight sandstone reservoirs in the Bozi gas reservoir of Kuqa depression was realized by using BP neural network. The results show that BP neural network is an effective method for predicting in-situ stress, with just a smaller error between the predicted and measured in-situ stress values, which is generally less than 10%. In the Lower Cretaceous of the Bozi gas reservoir, the maximum horizontal principal stress is the largest, followed by the vertical principal stress, and the minimum horizontal principal stress is the smallest. Generally, the strike-slip stress mechanism is universal, and the maximum horizontal principal stress is predominantly NE-SW trending in the east and NW-SE trending in the west of the Bozi gas reservoir. Compared with sandstone intervals, the sandstone-mudstone interbedding interval exhibits a stronger fluctuation of in-situ stress and a sudden high in-situ stress locally. The horizontal wells drilled along NE-SW direction in the east of the Bozi gas reservoir and along NW-SE direction in the west are relatively stable, while the vertical wells are prone to collapse.

Key words: Kuqa depression, Bozi gas reservoir, Cretaceous, ultra-deep formation, tight sandstone, present-day in-situ stress, BP neural network