A multi-layer perceptron neural network model for predicting the hydrate equilibrium conditions in multi-component hydrocarbon systems

This work presents a model based on multilayer perceptron neural network (MLPNN) for the prediction of hydrate equilibrium conditions in hydrocarbon systems. The model heuristics are based on an extensive experimental dataset found in the open literature (consisting of 2883 data points for pure component, 993 data points binary component and 484 data points for multicomponent systems, for a wide range of temperature, compositions, and considering different equilibrium phases and presence of inhibitors). Absolute average relative deviation (AARD), mean squared error (MSE) and the regression coefficient (R ²) are used as the evaluation criteria to test the efficacy and accuracy of the model’s performance. Results were validated with data points not used to develop the proposed model and found to be in close agreement. The model’s performance was also compared to well-known rigorous equilibrium models (Ng–Robinson and Colorado School of Mines models) and found superior in terms of accuracy with a AARD value as low as 0.60 MPa for the same experimental dataset. The results and comparison indicate that the proposed MLPNN model can be confidently used to predict hydrate equilibrium conditions for various hydrocarbon systems.