Study on development technology and policy of highly deviated wells in low-permeability oil reservoirs based on machine learning and numerical simulation

Abstract

This paper takes low permeability reservoir in Fang 854 block Chang 6 of Yanchang Oilfield as the object, and studies the development technology policy of highly deviated wells. The fine geological model is built by geostatistics and clustering algorithm, and the proxy model of numerical simulation is built by XGBoost machine learning algorithm, which realizes the second-level high-precision prediction of engineering parameters to development indicators. The field test shows that the strong heterogeneity of reservoir leads to significant differences in fracture morphology. The initial displacement intensity is controlled at 3.5 by implementing staged differential pressure fracturing technology, which can effectively inhibit sand production and extend the stable production period to 7-8 months. Combining multi-objective optimization algorithm with Monte Carlo simulation, this study established an integrated linkage model of "geology-engineering-economy". The results show that the average single well net present value of highly deviated wells deployed based on the data-driven optimization scheme reaches 6.05 million yuan, and the internal rate of return is 13.9%. This study shows that the introduction of machine learning and big data analysis can effectively break through the limitations of traditional trial and error method and provide scientific support for efficient development of low permeability complex reservoirs.