Machine Learning Prediction of Acid Fracture Performance in Naturally Fractured Dolomite Formations

Estimating the performance of acid fracture in naturally fractured formation is a complex and computationally expensive process. Therefore, machine learning was utilized in this research to simplify a complex physics-based acid fracture model in naturally fractured dolomite formations. The physics-based model integrates fracture propagation, reactive transport, heat transfer, and reservoir productivity model. The performance of acid fracture depends strongly on acid design parameters (i.e., injection rate, acid concentration, etc.) and reservoir/natural fractures properties (i.e., reservoir permeability, natural fractures spacing, etc.). A parametric study was conducted, and based on a statistical analysis of 4900 simulation scenarios, we found that the acid volume, reservoir permeability, and natural fracture spacing are the most significant parameters. Also, the injection rate and fluid rheology are of paramount importance. A model based on machine learning tools such as support vector machine, fuzzy logic system, and Artificial neural network (ANN) was developed. The artificial neural network model outperforms the other two tools and has a prediction error of 3.3%. Finally, an empirical equation has been obtained from the ANN model.