prediction of nigerian crude oil viscosity using artificial neural network

More documents

Recommendations

Info

186 highest correlation coefficient of 0.9989 as compared to those obtained using other correlations, clearly demonstrated that the ANN µ ob model were better predictors of viscosity than all the empirical correlations as graphically illustrated in Figures 1, 2, 3 and 4. Table 1: The results for the data used for the testing phase of μ ob mModel R s 1042 1119 977 1237 210 3100 1232 1137 1053 950 R Gasgra s API P μ μ μ ob obnn obchew − connally μobbeg− robinson 219 183 188 180 154 215 216 212 216 214 1.118 0.920 0.870 0.940 0.73 1.048 0.78 0.78 0.91 0.77 19 40.6 39.4 43.4 23.3 45.4 43.8 44.1 38.2 39.8 2951 2736 2858 2655 1680 4340 3963 3710 2965 3614 0.28 0.3 0.3 0.3 3.84 0.15 0.22 0.23 0.29 0.25 1.2 0.28 0.32 0.24 4.71 0.54 0.21 0.22 0.29 0.28 1.03 0.28 0.3 0.24 3.5 0.14 0.21 0.22 0.28 0.27 Table 2: The Statistical Analysis for μ ob Model Correlation Chew-Connally Begg-Robinson Neutral Network Ave. absolute relative error (E ave ) 0,45852 0,1741 0,06781 Minimum absolute relative error (E min ) 3,8 3,12 0,12532 Maximum absolute relative error 0,225806 0,391304 0,347826 (E max ) correlation coefficient (R 2 ) Fig. 3 Comparison of the estimated crude oil viscosity versus experimental crude oil viscosity (µ ob ANN model). Fig. 4 Comparison of the estimated crude oil viscosity versus experimental crude oil viscosity (Chew-Connally correlation). Fig. 5 Comparison of the estimated crude oil viscosity versus experimental crude oil viscosity (Beggs-Robinson correlation). Fig.6 Comparison of different correlation coefficients for different correlations of µ ob
187 9. Conclusion The newly developed back propagation neural network (BPNN) model for predicting crude oil viscosity was better than the empirical correlations. The µ ob ANN model, achieved the lowest average absolute relative error of 0.06781 and the highest correlation coefficient of 0.9989 as compared to existing empirical correlations. From the scatter diagrams for the ANN models and empirical correlations against their experimental values, the ANN models data points performed better than the existing empirical correlations. Nomenclature PVT Pressure, Volume, Temperature µ ob Crude oil viscosity at the bubble point P b Bubble point pressure, psia. R s Gas solubility, scf/stb T f Reservoir temperature, F API Oil gravity γ g (G.G) Gas gravity γ o Specific oil gravity E ave Average absolute relative error E min Minimum absolute relative error E max Maximum absolute relative error ANN Artificial neural network W ji Weight between neuron (i) in the layer (I) and neuron (j) in the layer(J) W ni Weight between neuron (i) in the layer (J) and neuron (n) in the layer (k) α Momentum factor E Error δ Derivative of E with respect to Ipk = Local Error η Learning rate constant BPNN Back propagation neural network RBFNN Radial basis function neural network PE Processing element X pi Input pattern in the input layer Subscript P, I , J , and K layers Ji between layers i and j Ni between layers j and k f k (I pk ) Sigmoid Transfer Function Derivative Acknowledgement The authors would like to extend their appreciation to our funding partners, The Petroleum Technology Development Fund, Nigeria whose contributions made this work possible. References [1] Deng, A. Deng:“ Prediction of PVT Oil Properties Using Artificial Neural Network”, M.Sc Thesis, University of Ibadan, Department of Petroleum Engineering, Ibadan, Nigeria, April 2007. [2] Bradley,H.B.: “Petroleum Engineering Handbook” ,SPE ,Richardson. [3] Tarek, A.H.: “Reservoir Engineering Handbook”, Gulf Publishing Company, Houston Texas 77252. [4] Tarek , A.H : “ Hydrocarbon phase Behaviors” , Houston, Gulf Publishing company. [5] Goda,H.M , Shokir,E.M , Fattah, K.A and Sayyouh,M.H: “Prediction of the PVT Data Using Neural Network Computing Theory” , SPE 85650, August 2003. [6] Osman,E.A., Abdel-Wahab,O. A and Al-Marhoun,M.A :“Prediction of Oil PVT Properties Using Neural Networks” , SPE 68233, March 2001.
Page 1 and 2: Petroleum & Coal ISSN 1337-7027 Ava
Page 3 and 4: 183 Where W ji is the weight of the
Page 5: 185 bubble point pressure (1420- 48

prediction of nigerian crude oil viscosity using artificial neural network

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?