摘要
页岩气集输管道在生产过程中的介质含砂量明显高于常规天然气且伴随着压力和流量的显著衰减,这导致集输管道管壁受到冲蚀迅速减薄甚至穿孔,为了实现对页岩气集输管道长期运行的可靠性评估和风险预警,采用了贝叶斯正则化(BR)方法训练的BP神经网络,建立了页岩气集输管道的壁厚动态预测模型。首先分析确定预测模型输入参量;然后通过管线对应数据进行BP神经网络的BR法训练及检验,并建立壁厚动态预测模型;最后通过该模型对实际监测数据进行实例应用,并与其他神经网络建立的预测模型相对比。该模型弥补了常规传统方法训练下BP神经网络模型容易陷入局部极小值的缺点,增强了模型的推广泛化能力,减小了壁厚预测误差提高了精度,并且能够根据新的监测数据进行动态预测,通过壁厚的预测来反映具体冲蚀情况。
The amount of sand produced in the production process of shale gas gathering and transportation pipelines is significantly higher than that of conventional natural gas and is accompanied by significant pressure and flow attenuation,which leads to rapid thinning and even perforation of the walls of gathering and transportation pipelines.In order to realize the long-term reliability assessment and risk warning of shale gas gathering and transportation pipelines,a BP neural network trained by Bayesian regularization(BR)method was used to establish a dynamic prediction model for the wall thickness of shale gas gathering and transportation pipelines.First analyzed and determined the input parameters of the predictive model.Then used the corresponding data to train and tested the BR method of the BP neural network,and established a dynamic prediction model.Finally,the actual monitoring data was applied to the actual monitoring data through this model,and compared with the prediction models established by other neural networks.This model makes up for the shortcomings of conventional BP neural network prediction models that tend to fall into local minimums,enhances the generalization ability of the prediction model,reduces the wall thickness prediction error and improves the accuracy,and is able to make dynamic predictions based on new monitoring data.And through the prediction of wall thickness to reflect the specific erosion situation.
作者
何跃
兰永福
樊星
贾文龙
冷翔宇
吴暇
黄军
韩西成
HE Yue;LAN Yong-fu;FAN Xing;JIA Wen-long;LENG Xiang-yu;WU Xia;HUANG Jun;HAN Xi-cheng(Downhole Operation Company Storage Reconstruction Research Center of CNPC Western Drilling Engineering Company,Karamay 834099,China;Petroleum Engineering School,Southwest Petroleum University,Chengdu 610500,China;Hanzheng Testing Technology Company,Guanghan 618399,China)
出处
《全面腐蚀控制》
2022年第10期13-20,共8页
Total Corrosion Control
关键词
BP神经网络
贝叶斯正则化
页岩气集输管道
壁厚动态预测
BPNN
bayesian regularization
shale gas gathering pipeline
dynamic prediction of wall thickness