The integrated geosciences and engineering production prediction in tight reservoir based on deep learning

Abstract

Listen

Translate article

The production prediction method of volume-fractured wells in tight reservoirs is studied. Different from the traditional fracture modeling description and the fracture seepage capacity description with analytical or semi-analytical methods, this study combines the concept of integrated geosciences and engineering, uses the simplified static data composed of wellbore data and dimensionality reduced formation data, with the daily data collected during the production process. The time series prediction models based on deep learning are adopted, and human interference such as well shut-in in the production process is considered, thereby increasing the accuracy of the calculation results and reducing the impacts of human factors in the time series prediction. Using time series prediction algorithms such as the Transformer model, the liquid production data of 56 wells in tight reservoirs in 4 formations were predicted, and the MSE reached 0.01957, indicating that the production prediction accuracy reached 98.043%, which is instructive to the production prediction in the actual production process.