Neural Network based Mechanical Earth Modelling (MEM): A case study in Hassi Messaoud Field, Algeria

Abstract

Accurate estimation of in-situ stresses is of great importance in the oil and gas industry from the exploration to the field development and production phases. The collected logs and mini-frac data in the last 15 years in the Hassi Messaoud Field (HMD), Algeria, shows that the reported state of stresses in this field is not consistent. This called for further studies to estimate more accurately the state of stresses, which more specifically will be used for the design of hydraulic fracturing at a later stage. This paper presents the results of the Mechanical Earth Model (MEM) constructed for the HMD field in Algeria. The results of the MEM, which is continuous logs of formations' elastic and strength properties, as well as the state of the in-situ stresses will be the direct input to the hydraulic fracture (HF) design. Hydraulic fracturing is the prime technique used in the field in order to enhance recovery from this tight sandstone reservoir. MEM is built beyond the conventional correlation with both a Generalized Linear Model (GLM), which serves as a simplified “ground truth” and three Artificial Neural Networks (ANN) layers: the first one is training mechanical properties from lab data, the second one trains shear and compressional sonic waves from acquired log data and the last one uses mini-frac data to calculate and train stress regimes from a poro-elastic model. The model is calibrated with the observed breakouts and fracturing data and a close agreement is observed. The work presented in this paper elaborates a detailed roadmap to accurately approximate and synthesize missing data using ANN's and extend it across the rest of the field to build, a 3D MEM. It is hoped that the results of this study can improve the HF operations in the field, which is currently reported to be not efficient.