Prediction of CO2–oil molecular diffusion using adaptive neuro-fuzzy inference system and particle swarm optimization technique

• ANFIS approach is implemented for accurate predicting the carbon dioxide diffusivity in oils.
• PSO algorithm is employed for obtaining the optimum parameters of the ANFIS model.
• The presented ANFIS model reproduces the corresponding data with high accuracy.

The quantification of carbon dioxide (CO2) dissolution in oil is crucial in predicting the potential and long-term behavior of CO2 in reservoir during secondary and tertiary oil recovery. Accurate predicting carbon dioxide molecular diffusion coefficient is a key parameter during carbon dioxide injection into oil reservoirs. In this study a new model based on adaptive neuro-fuzzy inference systems (ANFIS) is designed and developed for accurate prediction of carbon dioxide diffusivity in oils at elevated temperature and pressures. Particle Swarm Optimization (PSO) as population based stochastic search algorithms was applied to obtain the optimal ANFIS model parameters. Furthermore, a simple correlation is proposed for the application of interest. Although the prediction performance of regression model is high, the ANFIS model optimized by PSO algorithm exhibits better performance with average absolute relative deviation of 1.7% and squared correlation coefficient of 0.9987. Results from this study reveal that the proposed techniques can predict the CO2 molecular diffusion in oil with high accuracy. The tools developed in this study can be of immense practical values for experts and engineers to have a quick estimation on CO2 diffusion into reservoir oil at various conditions.