Application of artificial neural network for prediction of 10 crude oil properties

This study aims to develop an industrially reliable and accurate method estimate crude oil properties from their Fourier transform infrared spectroscopy (FTIR) spectra. We used the complete FTIR spectral data of selected samples seven different Canadian fields predict 10 important using artificial neural networks (ANNs). The predicted include specific gravity, kinematic viscosity, total acid number, micro carbon content, production light heavy naphtha, Kero, distillate in refineries. 107 (65 42 heavy/medium samples) this came reservoirs across Canada. In line with standard practice, we 80% dataset for training ANN models remaining 20% test models. analysis, mean squared error (MSE) was as loss function models, absolute prediction (MAPE) a reference compare performance constructed numbers layers. work demonstrates that is promising technique provides rapid estimates interest industry. A comparison values by validated corresponding measured (actual) showed excellent acceptable range (below 15%) aimed our industry partner all except which building based on natural logarithmic viscosities significantly improved results.