Comparing ANN with PSO-ANN in Prediction of Natural Gas Hydrate Formation Parameters

This study explores the application of artificial intelligence models, specifically an artificial neural network (ANN) and a hybrid Particle Swarm Optimization-ANN (PSO-ANN), to predict the hydrate formation temperature (HFT) of natural gas hydrates. The research focuses on the impact of mole fractions of gas components and system pressure on HFT. Utilizing 1841 experimental data samples, with 80 % for training and 20 % for testing, the models were optimized through trial-and-error. Performance metrics, including mean squared error (MSE) and coefficient of determination (R2), were used to evaluate the models’ accuracy. The ANN model achieved MSE and R2 values of 0.0003533 and 0.9976, respectively, while the PSO-ANN model yielded MSE and R2 values of 0.0003298 and 0.9899. Both models demonstrated high accuracy, with the PSO-ANN model showing a slight edge in MSE. While traditional EOS models and previous ANN-based studies have shown the ability to predict HFT, they often face challenges with accuracy and generalizability, particularly in multicomponent systems. This study aims to address these challenges by demonstrating the efficacy of a streamlined ANN model, optimized using PSO, that focuses on the most critical input parameters to achieve high accuracy and generalization. Our approach simplifies the modeling process by reducing the need for extensive input variables, thereby addressing limitations observed in prior studies.