Development and Evaluation of Deep Learning Models for Forecasting Gas Production and Flowback Water in Shale Gas Reservoirs

Abstract

In this paper, deep learning models are developed based on a multilayer perceptron to forecast 12 month cumulative produced shale gas and 90 day produced flowback water using a study area within the Eagle Ford Formation as a database. These models can help decision-makers have important references when drilling new wells. Latitude, longitude, true vertical depth, lateral longitude, total proppant, and total fracture water are used as input variables. The trained models are evaluated in a study area within the Burgos Basin to analyze the energy benefits, freshwater consumption, and flowback water produced through well drilling for the first time in Mexico. A non-dominated sorting genetic algorithm is used for the minimization of total freshwater consumption and maximization of cumulative produced gas. The multilayer perceptron models showed good results. The coefficient of determination and the mean absolute error of the single-output models remained above 0.93 and below 3.00 for the validation set, respectively.

Process Systems Engineering Tools for Optimization of Trained Machine Learning Models: Comparative and Perspective

“…Their ability to mimic the behavior of more complex systems is transforming how engineers approach process design, analysis, and optimization. The advancement and implementation of data-driven surrogate models in PSE have been thoroughly recognized. − Moreover, data-driven surrogate models represent a significant improvement in accuracy and performance in the modeling of complex processes, including those that were not previously able to be addressed. The abilities of these surrogate models have led to more precise results, which in turn improves decision-making and benefits sustainability.…”