Modeling asphaltene precipitation during natural depletion of reservoirs and evaluating screening criteria for stability of crude oils

Abstract

Asphaltene precipitation and deposition can cause serious operational problems during oil production, transportation and processing. This mandates careful monitoring of crude oil stability by studying various thermodynamic and structural parameters contributing to crude oil instability. This study aimed at application of various optimized intelligent techniques to forecast the amount of asphaltene precipitation. A large database was prepared using 24 different Iranian crude oils with asphaltene precipitation problem during natural depletion of the reservoirs. Pressure, temperature, bubble point pressure, Oil API gravity and Saturate-Aromatics-Resin-Asphaltene (SARA) fractions were selected as the input parameters of all models. Two intelligent techniques; namely Radial Basis Function (RBF) and Multilayer Perceptron (MLP) neural network were developed, and various optimization techniques; including Genetic Algorithm (GA), Gravitational Search Algorithm (GSA), Imperialist Competitive Algorithm (ICA), Particle Swarm Optimization (PSO), Differential Evolution (DE), Ant Colony Optimization (ACO), Scaled Conjugate Gradient (SCG), Levenberg-Marquardt (LM), Resilient Back Propagation (RB), and Bayesian Regularization (BR) were applied. Graphical and statistical error analyses suggest that RBF method optimized with PSO algorithm and MLP method optimized with BR, provide the best performance. Moreover, a correlation was developed to estimate the amount of asphaltene precipitation using gene expression programming (GEP). The output reflects excellent capability of the correlation to estimate the experimental results with good accuracy, however to a lesser extent than intelligent modeling. In addition, the predictions from the current models were compared to those of Flory-Huggins thermodynamic model. The results confirm that the current models are more accurate and computationally less expensive than Flory-Huggins model. Moreover, Colloidal Instability Index (CII) and Asphaltene to Resin ratio (A/R) were applied as indicators to forecast the instability of crude oil. It was found that CII could successfully estimate the instability of 20 crude oils, while A/R ratio only estimated the instability of 10 crude oils accurately.