Implementation of artificial intelligence and support vector machine learning to estimate the drilling fluid density in high-pressure high-temperature wells

Abstract

One of the challenging conditions in wellbore management is high-pressure, high-temperature wells that apply large expenditures and maintenance costs to petroleum industries. In this regard, drilling fluid rheology and its crucial factors would help engineers to have a fundamental understanding of the most appropriate wellbore management. Moreover, it can help engineers to control the fluid loss phenomenon. This paper aimed to consider different artificial intelligence and machine learning models to predict the drilling fluid density and select the optimum model, which can be extended to field applications. These models are ANFIS (adaptive neuro-fuzzy inference system), PSO-ANFIS (particle swarm optimization-adaptive​ neuro-fuzzy inference system), LSSVM-GA (least square support vector machine-genetics algorithm), and RBF (radial basis function) algorithm were modeled in the programming era. In LSSVM-GA model, it is evident that the proper linear equation for the prediction of drilling fluid density is “y = 1.0041x + 0.0019” with the correlation factor of (R2=0.9966). Due to the advantages of RBF algorithm to other genetics algorithm, this algorithm was used in this part to predict the drilling fluid density. Therefore, it is evident that the proper linear equation for the prediction of drilling fluid density is “y = 1.0009x + 0.0034” with the correlation factor of (R2=0.9999). Consequently, RBF model was selected as the optimum method in drilling fluid prediction. Furthermore, there is a proper match with training and experimental data and the maximum deviation is about -0.006. In ANFIS algorithm, exponential derivation would be preferred to predict the drilling fluid density instead of linear equation. Consequently, the RBF model provide has a good agreement with the experimental drilling fluid density. It is indicated that the model have appropriate accuracy and validity with experimental data.