基于KPCA-IGOA-ELM的油气混输管道腐蚀速率预测模型  

Corrosion rate prediction model of oil-gas mixed transportation pipelines based on KPCA-IGOA-ELM

作　　者：吕林林[1] 王杰 祁庆芳 郭策 贺蓉蓉 孙小伟 LYU Linlin;WANG Jie;QI Qingfang;GUO Ce;HE Rongrong;SUN Xiaowei(Pipeline Office of PetroChina Qinghai Oilfield Company)

机构地区：[1]中国石油天然气股份有限公司青海油田分公司管道处,青海省海西蒙古族藏族自治州茫崖市816000

出　　处：《油气储运》2023年第7期785-792,共8页Oil & Gas Storage and Transportation

基　　金：中国石油天然气股份有限公司重点项目“青海油田涩北气田腐蚀机理研究”,2022-KJ-105。

摘　　要：油气混输管道内腐蚀速率较大,准确预测混输管道内腐蚀速率对于提升管道完整性管理水平具有重要意义。针对这一问题,首先,利用现场监测结果构建评价指标体系和数据集,采用核主成分分析法(Kernel Principal Component Analysis,KPCA)进行降维操作,随后利用改进的蝗虫算法(Improved Grasshopper Optimization Algorithm,IGOA)对极限学习机(Extreme Learning Machine,ELM)进行优化,确定最优网络结构和激励函数,提出了KPCA-IGOA-ELM组合预测模型。利用该模型,以8组数据为基础进行预测,并与其他模型预测结果进行对比,以此验证预测效果。结果表明:KPCA算法共提取出3个主成分,简化了ELM模型的网络结构,其中H_(2)S分压、CO_(2)分压、Ca^(2+)浓度、Cl^(-)浓度、温度、流速对腐蚀作用的贡献较大;通过试算法确定最优ELM模型的网络结构为3-32-1,激励函数为Sigmoid函数,此时的均方根误差最小;KPCA-IGOA-ELM组合预测模型的均方根误差、平均绝对百分误差、希尔不等系数分别为0.00256、2.45834、1.113,平均训练时间为4.19 s,均优于其他模型。对于油气混输管道,KPCA-IGOA-ELM模型是一种较为优秀的算法,可在实际中推广应用。The oil-gas mixed transportation pipeline has a high internal corrosion rate.Hence,accurately predicting the internal corrosion rate of mixed transportation pipelines is of great significance to improve the integrity management of pipelines.In response to this problem,the evaluation index system and data set were constructed based on the field monitoring results at first,and the Kernel Principal Component Analysis(KPCA)was used for dimensionality reduction.Then,the Improved Grasshopper Optimization Algorithm(IGOA)was adopted to optimize the Extreme Learning Machine(ELM),the optimal network structure and the excitation function were determined,and the combined prediction model of KPCA-IGOA-ELM was proposed.With this model,prediction was performed based on 8 groups of data,and comparison was made with the results of other models to verify the prediction results.The results showed that three principal components were extracted by KPCA,and the network structure of ELM model was simplified.Among them,H_(2)S partial pressure,CO_(2)partial pressure,calcium ion,chloride ion,temperature and flow rate have significant contribution to corrosion.In addition,the network structure of the optimal ELM model was determined as 3-32-1 by trial algorithm,and the excitation function was the Sigmoid function,having the minimum root mean square error.Moreover,the RMSE,MAPE and Theil IC of the KPCA-IGOA-ELM combined prediction model are 0.00256,2.45834 and 1.113,respectively,and the average training time is 4.19 s,which are all superior to other models.It is proved that the KPCA-IGOA-ELM model is an excellent algorithm for oil-gas mixed transportation pipelines,which could be popularized and applied in practice.

关 键 词：油气混输 腐蚀预测 KPCA-IGOA-ELM 算法优化 影响因素 

分 类 号：TE832[石油与天然气工程—油气储运工程]