考虑多参数关联融合的井口流量深度预测模型  

Wellhead Flow Depth Prediction Model with Multi-parameter Correlation and Integration

作　　者：张碧波 廖华伟 刘欢 兰辉 ZHANG Bibo;LIAO Huawei;LIU Huan;LAN Hui(PetroChina Southwest Oil and Gas Field Chuandongbei Operation Branch,Chengdu 610020,China)

机构地区：[1]西南油气田公司川东北作业分公司,成都610020

出　　处：《自动化与仪表》2025年第4期67-71,共5页Automation & Instrumentation

摘　　要：针对高含硫气体容易造成传统的单井计量物理仪表(如差压式流量计、涡轮流量计)偏差、损坏、保养困难等难点,提出了一种基于多参数关联融合的循环神经网络井口流量预测模型,通过单个变量和多个变量的分析,以期获得合理有效的井口流量预测结果。设计LSTM网络结构,研究单变量LSTM井口流量预测模型和多变量LSTM井口流量预测模型。研究以川东北作业公司的LJ-24井为例,建立了单变量和多变量神经网络井口流量预测模型并对某一个月的井口流量数据进行预测,取得了较好的实际应用效果。Aiming at the difficulties that high-sulfur gas can easily cause deviation,damage,and maintenance difficulties of traditional single-well metering physical instruments(such as differential pressure flowmeters,turbine flowmeters),this paper proposes a recurrent neural network wellhead flow prediction model based on multi-parameter association fusion.Through the analysis of single variables and multiple variables,it is hoped to obtain reasonable and effective wellhead flow prediction results.The LSTM network structure is designed to study the single variable LSTM wellhead flow prediction model and the multivariate LSTM wellhead flow prediction model.This study takes the LJ-24 well of Chuandongbei operation branch as an example.Establish a univariate and multivariate neural network wellhead flow prediction model and predict the wellhead flow data in a month,achieving good practical application results.

关 键 词：井口流量预测 虚拟计量 LSTM 单变量 多变量 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]