Boundary prediction method of gas hydrate risk prevention and control based on 1DCNN-SVM

To ensure the safe flow of oil and gas pipelines, accurate prediction of the formation of natural gas hydrates is crucial. Traditional prediction methods rely on experimental empirical formulas or simple physical models, being complex in computation, thus have limited applicability, and poor accuracy. In this study, an innovative prediction method based on onedimensional convolutional neural network(1DCNN)-suppoi vector machine(SVM)for the phase equilibrium of natural gas hydrates was proposed. In the experiments, the impact of different iteration numbers on model performance was explored, and it was determined that the model performed best after 2 000 iterations. A comparison of the 1DCNN-SVM model with traditional SVM, CNN, and BP neural network models showed that the 1DCNN-SVM model exhibited superior prediction performance, with an R² of 0.976 1, MSE of 1.823 6, and MAE of 0.588 9, all outperforming the compared models. Besides, the 1DCNNSVM model demonstrated strong applicability and stability when applied to new data. This study verifies the feasibility of using the 1DCNN-SVM model for predicting natural gas hydrate formation conditions, giving a new approach for the prediction, monitoring, early warning, and prevention of hydrate formation in oil and gas pipelines.