During oil processing, the viscosity can change several times as the fluids move through reservoirs to pipelines and experience phase separation or changes in temperature, pressure, and composition. Therefore, reliable and accurate data about fluid viscosity in a wide range of conditions and combinations is very important.In this research, it has been tried to take innovation in the optimization of models which was developed in the past by researchers and scientists , such as Friction Theory (FT) and Free Volume Theory (FVT) models coupling with two and three parameter cubic state equations which include Redlich-Kuang (SRK), Peng-Robinson (PR) , Patel-Teja (PT), and PC-SAFT equation of state as a model of the family of statistical methods with new optimization algorithms including CMEAS algorithm, Active-set, SQP,Fminsearch, GA and PSO to provide the best model for predicting the viscosity of pure, two-component, multi-component hydrocarbons and finally the viscosity of crude oil , providing a comparison between the prediction results of all models.The input data of this research includes SARA test results for 15 crude oil samples from the research of Faraji et al., as well as 7 crude oil samples from the research of Abutaqieh et al. Multiple components data at different pressures and temperatures have been adapted from various literature references. 70 % of the data have been eva‎luated for training the algorithm and thirty percent of the data have been used as validation.The results of this research show that in the case of pure hydrocarbons, with an average correlation index of R2=0.991%, no significant difference was observed between the prediction models, only in some compounds, the free volume model showed more error than the rest of the models. From the two models of friction theory and free volume theory, FVT model has more error compared to FT.Patel-Teja cubic state equations provided higher prediction accuracy than other equations with average validation index = 0.67% SMAPEv, which had not been investigated in any article, also PC-SAFT equation in both FT and FVT models with average Validation index = 0.80% SMAPEv showed the lowest prediction accuracy,which is also very significant.In the case of binary compounds, the results of the research show that in the case of heavy compounds, PT-FT-SQP prediction model showed higher accuracy than other models with validation parameter SMAPEv=0.64%. Also, the results indicated that in binary compounds and In case of multicomponent viscosity prediction, FT model is highly far from the FVT model in forecasting, and this can be attributed to the greater density and weight of the compounds compared to pure hydrocarbons. In the case of crude oil, very interesting results were obtained, which show that the best forecasting model in terms of the residual model of the friction theory, in terms of the equation of state, it was the PC-SAFT equation with an average validation index = 2.26% SMAPEv and Patel Teja validation index = 2.23% SMAPEv, and in terms of the optimization algorithm, the SQP algorithm with a validation index = 2.53% SMAPEv. The results show that the percentage of errors for predicting the viscosity of crude oil is higher than that of pure and multicomponent compounds, which can be related to the complexity and non-ideality of crude oil composition, which is still a special challenge for optimization.

حميدرضا شاه وردي

محسن محمدي

روح اله هاشمي , عليرضا خزعلي