摘要
停输操作在福建成品油管道运行管理中时常发生,停输再启动过程中管内压力变化剧烈。当管内压力下降时,现场人员经常误以为发生泄漏等异常事故,需巡检排查,增加了现场的管理难度。为了提高现场管理水平,从机理模型分析了管道停输时段压力变化的影响因素,并基于出站油温、压力等检测数据,同时考虑环境温度的变化,建立基于机器学习算法的压力预测模型,监测管内压力变化情况。以泉港南线和黄塘溪东出站点为例,以RMSE、MAE、R2为指标,对比了LR、SVM、DT、RF、GB这5种回归预测模型。结果表明:DT、RF、GB模型适用于福建成品油管道停输的管内压力变化分析,而RF准确度最高。建立预测模型后,利用天气预报的气温数据,可以预测未来时段管内压力的变化趋势,当预测值与检测值差别较大时,实现报警功能。
Shutdown often occurs in the operation and management of Fujian oil product pipeline,and the pipeline pressure changes drastically during the shutdown and restart.When the pipeline pressure drops,the on-site personnel often mistakenly believe that an abnormal accident like a leak has occurred,which requires inspections and increases the difficulty of on-site management.In order to im-prove the level of on-site management,the influencing factors of pressure changes during the pipeline shutdown period from the mechanism model are analyzed.Based on the detection data such as outbound oil temperature and pressure,taking into account changes in ambient temperature,a pressure prediction model based on machine learning algorithms is established to monitor the pipeline pressure changes.Taking the Quangang South Line and Huangtangxi East exit as examples,and taking RMSE,MAE,and R2 as indicators,five regression prediction models including LR,SVM,DT,RF,and GB are compared.The results show that the DT,RF,and GB models are suitable for the analysis of pipeline pressure changes of Fujian oil product pipeline,and the RF's accuracy is the highest.After the prediction models are established,by using the forecasting temperature data,the change trend of the pipeline pressure can be predicted.When the predicted value and the detected one are significantly different,the alarm function is realized.
作者
邵晓
郑坚钦
戴元豪
张扬
SHAO Xiao;ZHENG Jianqin;DAI Yuanhao;ZHANG Yang(Fujian Petroleum Branch,SINOPEC;China University of Petroleum(Beijing))
出处
《油气田地面工程》
2020年第10期68-72,共5页
Oil-Gas Field Surface Engineering
关键词
成品油管道
停输
压力变化
机器学习
预测模型
product oil pipeline
shutdown
pressure change
machine learning
prediction mode
