Artificial Neural Network for Prediction of Hydrate Formation Temperature

Gas hydrate deposition is one of the major Flow Assurance problems in petroleum production in the offshore environment. Therefore, is important to accurately predict hydrate formation conditions and avoid these conditions or propose a hydrate management plan. This study compares the effectiveness of Artificial Neural Network (ANN) for predicting hydrate formation temperature to the effectiveness of other hydrate temperature prediction correlations such as: Towler and Mokhtab correlation, Hammerschmidt correlation and Bahadori and Vuthalaru correlation. The ANN was trained using 459 hydrate formation experimental data points from Katz chart and Wilcox et al chart. Pressure (P) and specific gravity (ϒ) were chosen as the inputs in the 4-layer network while temperature was the output. The data points were for gases of specific gravity of 0.5539, 0.6, 0.7, 0.8, 0.9 and 1.0. The experimental pressures considered were from 49 psia to 4000 psia. The Neural Network was built using an excel add-in tool, NEUROXL. ANN accurately predicted the experimental hydrate formation temperature with the regression coefficient greater than 0.98 for the different specific gravities considered. Moreso, the error analysis shows ANN performed better than Towler and Mokhtab correlation, Hammerschmidt correlation and Bahadori and Vuthalaru correlation because it had the least Mean Absolute percentage error, MAPE (3.5) compared to the other correlations. ANN is a viable tool for hydrate prediction and the current model can be improved upon by including more experimental data in the ANN.