摘要
目前,我国大多数油气管道服役时间已超过20 a,受到土壤、大气以及水体等多方面的影响,腐蚀成为危害管道安全、造成管道失效的重要因素。针对单一腐蚀缺陷管道剩余强度样本数据少、公式计算保守性强、有限元分析过于复杂等缺点,将RS、PSO和SVM算法模型有机结合,构建了腐蚀管道剩余强度预测模型。通过RS属性约简,有效提取了影响管道剩余强度的关键性指标因素,随后应用改进的PSO算法对SVM的参数进行了寻优,避免了人工试算法造成的误差过大和训练时间过长的缺点,与BP神经网络、RS-WNN算法相比,RS-PSO-SVM算法的保守性和准确性都较为优越,平均绝对百分误差为1.23%,均方根误差为0.17 MPa,模型的鲁棒性和预测性更好,对管道剩余强度的研究具有借鉴意义。
At present,the service time of most oil and gas pipelines in China has exceeded 20 years.Due to the influences of soil,atmosphere,water and so on,the pipeline corrosion has become an important factor endangering pipeline safety and causing pipeline failure.In view of the shortcomings such as less residual strength sample data of single defect corrosion pipeline,strongly conservative formula calculation,and too complex finite element analysis,the RS,PSO and SVM algorithm models were organically combined to build the residual strength prediction model of corrosion pipeline.Through RS attribute reduction,the key indicators effectively affecting pipeline residual strength were obtained,then the improved PSO algorithm was applied to optimize the parameters of SVM,avoiding excessive error and too long training time caused by manual trial calculation.Compared with BP neural network and RS-WNN algorithms,RS-PSO-SVM algorithm is better in conservation and accuracy,its average absolute percentage error is 1.23%,root mean square error is 0.17 MPa,robustness and predictability are good.
作者
杨旭东
周艳丽
刘志娟
陆亮
于天齐
刘勇
YANG Xudong;ZHOU Yanli;LIU Zhijuan;LU Liang;YU Tianqi;LIU Yong(No.1 Oil Production Plant of Huabei Oilfield Company,PetroChina,Renqiu 062552,China)
出处
《石油工程建设》
2020年第3期8-12,共5页
Petroleum Engineering Construction
