Search

In order to mitigate the issues caused by hydrate blockage in flow assurance, experimental data on natural gas hydrate formation was collected to construct a Probabilistic Neural Network (PNN) model. By improving the Aquila Optimizer (AO) algorithm through adaptive weights and hyperbolic tangent function, optimization of smoothing parameters was achieved, resulting in the establishment of an IAO-PNN-based hydrate formation prediction model. A comparison with thermodynamic models and machine learning models validated the superiority of the algorithm. The results show that the improved AO algorithm (IAO) exhibits significantly higher optimization precision and convergence speed compared to intelligent algorithms such as AO, PSO and SSA. The IAO-PNN model exhibits the highest consistency with experimental data, making it suitable for predicting hydrate formation conditions in binary systems, multicomponent systems, acid systems and alcohol-salt systems, and it demonstrates good predictive performance in high- pressure environments. Compared to thermodynamic models and machine learning models, the IAO-PNN model shows excellent generalization performance, with an root mean square error (RMSE) of 0.6176 and an coefficient of determination (R²) of 0.9994 on the training set, and an RMSE of 0.7624 and an R² of 0.9991 on the test set. Through on-site verification, the IAO-PNN model displays good applicability and can provide reference for formulating on-site remediation measures. [ABSTRACT FROM AUTHOR]