基于深度学习的用气趋势预测与管网气量调配算法——以中国石油西南油气田公司为例

Deep learning-based prediction of gas consumption trend and algorithms on pipeline gas regulation:An example from PetroChina Southwest Oil&Gasfield Company

生产实时数据是油气生产的重要资产,通过监测实时数据,管理人员可以及时掌握油气田的生产运行状况,目前数据使用主要集中在数据监测、设备联锁、支撑报表自动生成等应用。为了提升已有数据价值、支撑调度管理,对某气矿已采集天然气开发生产实时数据再利用,通过深度学习进行用户用气量趋势预测,并研究出一种管网气量调配的算法,实现了管网输气调配方案的自动生成。研究结果表明:①通过对历史数据的处理,可以对异常值进行分析和过滤,从而获取能够使用分析的数据,并能生成参数的正常运行区间,对有效数据提取、指导工艺参数运行范围具有重要意义;②通过神经网络进行用户用气量小时趋势的预测,能较好地捕获未来一段时间内的用气量变化趋势,从而实现对需求量变化的提前感知;③通过构建最少调整次数和最接近标定产能两种基于管网汇总气量的井口产量调整算法,实现了气井生产管网调配方案的自动计算,减少了人工计算分析管网气量调配的工作量。

All the real-time production data is an important asset in oil and gas production.Managers can make clear the production and operation situations of oil and gas fields in time by means of data monitoring.At present,the data is mainly used for data monitoring,equipment interlocking and automatic generation of support reports.In order to enhance the value of existing data and support allocation management,one gas district collected and reused the real-time data to predict the trend of users'gas consumption on the basis of deep learning.In addition,an algorithm on pipeline gas regulation was used to realize the automatic generating a regulation scheme.Results show that(i)after historical data processing,exceptional data may be analyzed and screened out,so as to obtain available one and generate the normal data range,which is of great significance to extract available data and guide the range;(ii)neural network is adopted to predict the trend of gas consumption per hour,which may better capture another changing direction in the near future,so as to predict the change in gas demand as early as possible;and(iii)the establishment of two algorithms on wellhead production regulation based on sum pipeline gas,namely the minimum regulation times and the closest to calibrated productivity can realize the automatic calculating the regulation scheme and reduce the workload of manual calculation on pipeline gas regulation.