基于改进时序卷积网络的采油速度预测模型  

Oil Recovery Rate Prediction Model Based on Improved Temporal Convolutional Network

作　　者：张强[1] 邓彬 李志溢 袁和平 ZHANG Qiang;DENG Bin;LI Zhiyi;YUAN Heping(School of Computer&Information Technology,Northeast Petroleum University,Daqing 163318;The Fifth Oil Production Plant,Daqing Oilfield Limited Company,Daqing 163513)

机构地区：[1]东北石油大学计算机与信息技术学院,大庆163318 [2]大庆油田有限责任公司第五采油厂,大庆163513

出　　处：《计算机与数字工程》2025年第3期734-740,共7页Computer & Digital Engineering

基　　金：国家自然科学基金项目(编号:61702093);黑龙江省自然科学基金项目(编号:F2018003);黑龙江省博士后专项(编号:LBH-Q20077)资助。

摘　　要：采油速度是表征油田开发速度与能效的重要指标,是油田注采优化必不可少的研究内容。基于多种不确定性因素的干扰,论文提出一种改进时序卷积网络的采油速度预测模型。应用注意力机制(Attention)关注全局特征与局部特征的关系并为每个部分赋予不同权重,改善网络的学习能力。引入AR自回归组件为预测加入线性成分,提升模型对输入尺度变化的敏感度。为验证改进模型的有效性,选取传统时序卷积网络与长短期记忆网络(LSTM)、门控循环单元(GRU)及融入注意力机制的LSTM、GRU作为对比模型。实验结果表明,论文提出的模型具有更高的预测精度和更稳定的效果。Oil production rate is an important indicator to characterize oilfield development rate and energy efficiency,and is an essential research content for oilfield injection-production optimization.Based on the interference of various uncertain factors,this paper proposes an oil production rate prediction model of improved temporal convolutional network.Applying attention mechanism pays attention to the relationship between global and local features and assigns different weights to each part to improve the learning ability of the network.The AR autoregressive component is introduced to add a linear component to the prediction to improve the sensitivity of the model to changes in the input scale.In order to verify the effectiveness of the improved model,traditional temporal convolutional network and long short term memory(LSTM),gate recurrent unit(GRU)and LSTM,GRU integrated with attention mechanism are selected as comparative models.The experimental results show that the model proposed in this paper has higher prediction accuracy and more stable effect.

关 键 词：时序卷积网络 注意力机制 AR自回归组件 采油速度预测 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]