Using Machine Learning Method to Optimize Well Stimulation Design in Heterogeneous Naturally Fractured Tight Reservoirs

Abstract

Abstract The reservoirs in Kuqa foreland area of Tarim Basin in China are ultra-deep HTHP (High Temperature and High Pressure) naturally fractured sandstone reservoirs. Due to low permeability of the matrix (<0.1mD), stimulation of the natural fractures is the key to well productivity enhancement. Different stimulation techniques with different stimulation strengths have been tried in the last decade, but stimulation effectiveness varied. Therefore, machine learning method is employed to identify the main controlling factors and optimize the well stimulation design. Firstly, geological data, stimulation data, productivity data, etc. for more than 200 wells were used to develop data analysis models, and the major characteristic parameters and their weightiness were determined through machine learning. Afterwards, the stimulation parameters of these wells, including injection rate, fluid volume, proppant volume, etc., were correlated with post-stimulation open flow capacity increments using several regression modeling methods, and the weightiness of these stimulation parameters was determined through machine learning. Cross validation method was used to choose the most accurate and stable model, which was then used to optimize the stimulation parameters of new wells. The model is applied to two test wells. The stimulation technologies and stimulation parameters of the two wells are optimized. Compared with the natural productivity, the productivity after stimulation was increased by 5.5 times and 21.5 times respectively. Machine learning algorithms are used to find an implicit rule from a large amount of data and express the rule with a high dimension nonlinear algorithm equation. It is very useful but seldom has applications in the area of reservoir stimulation. This paper found the controlling parameters of reservoir stimulation in Kuqa foreland area of Tarim Basin through machine learning and successfully used it in well productivity enhancement practices.