摘要
为解决油气管道腐蚀预警成本高、精度低的难题,研究整合知识图谱与神经网络技术,开展腐蚀预测研究。通过BERT-BiLSTM-CRF算法构建专业知识图谱,利用Neo4j完成结构化存储;选取8项关键参数作为输入,对比3种神经网络构建基础预测模型,并引入粒子群和遗传算法进行优化迭代。结果表明,知识抽取算法的精准率、召回率和F1值分别达到94.56%,90.48%和92.47%,BP神经网络在基础模型中表现最优;经PSO优化后的PSO-BP模型性能显著提升,训练与测试决定系数高达0.995 0和0.994 3。该研究为油气管道安全运维提供了高效的解决方案。
To tackle the challenges of high costs and low accuracy in corrosion warning for oil and gas pipelines,this study integrates knowledge graph and neural network technologies to conduct corrosion prediction research.A professional knowledge graph is constructed using the BERT-BiLSTM-CRF algorithm,and Neo4j is employed for structured storage.Eight key parameters are selected as inputs,and basic prediction models are built by comparing three types of neural networks.Additionally,the particle swarm optimization(PSO)and genetic algorithm(GA)are introduced for optimization and iterative improvement.The results show that the precision,recall,and F 1-score of the knowledge extraction algorithm reach 94.56%,90.48%,and 92.47%,respectively.Among the basic models,the BP neural network performs the best.After optimization by PSO,the performance of the PSO-BP model is significantly enhanced,with the determination coefficients of the training and testing phases reaching 0.9950 and 0.9943.This study provides an efficient solution for the safe operation and maintenance of oil and gas pipelines.
作者
谢若涵
范峥
李珍
郝新宇
韩洁
李稳宏
XIE Ruohan;FAN Zheng;LI Zhen;HAO Xinyu;HAN Jie;LI Wenhong(College of Chemistry&Chemical Engineering,Xi’an Shiyou University,Xi’an 710065,China;Shaanxi Chemical Industry Research Institute Co.,Ltd.,Xi’an 710061,China;Shaanxi Yanchang Petroleum(Group)Company Limited Refining and Chemical Company,Yan’an 727406,China;Shaanxi Mingze Yisheng Energy Technology Company Limited,Xianyang 712000,China;School of Chemical Engineering,Northwestern University,Xi’an 710069,China)
出处
《应用化工》
北大核心
2025年第11期2894-2898,共5页
Applied Chemical Industry
基金
西安石油大学研究生创新与实践能力培养计划资助(YCS23214233)。
关键词
油气管道
均蚀速率
知识图谱
神经网络
优化
oil and gas pipeline
uniform erosion rate
knowledge graph
neural network
optimisation
