基于1DCNN-SVM的天然气水合物风险防控边界预测方法  

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

作　　者：吕晓方[1] 陈书楷 徐孝轩 柳扬 钱瑞祥 王传硕 李晓伟 周诗岽[1] LYU Xiaofang;CHEN Shukai;XU Xiaoxuan;LIU Yang;QIAN Ruixiang;WANG Chuanshuo;LI Xiaowei;ZHOU Shidong(Provincial Key Laboratory of New Energy Storage and Transportation Technology,Changzhou University;Institute of Petroleum Exploration and Development Research,SINOPEC)

机构地区：[1]常州大学·新能源储运技术省高校重点实验室,江苏省常州市213164 [2]中国石油化工股份有限公司石油勘探开发研究院

出　　处：《管道保护》2025年第1期14-21,共8页

基　　金：国家自然科学基金面上项目“深水浅层非成岩水合物浆体分解聚并/沉积机理与多相流动规律研究”(52274061)。

摘　　要：为了保障油气管道的流动安全,准确预测天然气水合物的生成条件非常重要。传统方法依赖于实验经验公式或简单物理模型,但这些方法计算复杂、适用范围有限且精度较低。为此,提出了一种基于一维卷积神经网络(1DCNN)-支持向量机(SVM)的天然气水合物相平衡预测方法。在实验中,探讨了不同迭代次数对模型性能的影响,确定2000次迭代时模型性能最佳。对比1DCNN-SVM模型与传统SVM、CNN、BP模型和OLGA的预测效果,结果显示1DCNN-SVM模型具有优异的预测性能,R2达到0.9761,MSE为1.8236,MAE为0.5889,均优于其他模型。此外,1DCNN-SVM模型在面对新数据时,表现出良好的适用性与稳定性。该预测方法为油气管道水合物生成的预测、监测预警及防控提供了新的思路。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 one-dimensional convolutional neural network(1DCNN)-support 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 2000 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 0.9761,MSE of 1.8236,and MAE of 0.5889,all outperforming the compared models.Besides,the 1DCNN-SVM 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.

关 键 词：天然气水合物 相平衡 一维卷积神经网络(1DCNN) 支持向量机(SVM) 

分 类 号：TE832[石油与天然气工程—油气储运工程]