基于长短期记忆神经网络算法的稠油井转周时机优化  

Optimization of Heavy Oil Well Rotation Timing Based on Long Short-Term Memory Neural Network Algorithm

作　　者：常峰[1] 宋清新[1] 李颖 赵云献[1] 侯宝宁[1] CHANG Feng;SONG Qingxin;LI Ying;ZHAO Yunxian;HOU Baoning(Sinopec Shengli Oilfield Branch,Dongying,Shandong,257000,China;Shandong Shengruan Technology Co.,Ltd.,Dongying,Shandong,257000,China)

机构地区：[1]中国石化胜利油田分公司,山东东营257000 [2]山东胜软科技股份有限公司,山东东营257000

出　　处：《自动化应用》2024年第20期119-121,共3页Automation Application

摘　　要：根据稠油井蒸汽吞吐的生产特点和胜利油田稠油井注汽、生产的历史数据,依据液量、含水分类构建样本库,采用长短期记忆网络(LSTM)算法建立稠油井转周时机预测模型,持续进行模型训练、优化,预测稠油井未来90天的生产指标,实现稠油注汽转周多维度智能预测,提高稠油井产量预测准确率、稠油注汽转周最佳时机预测准确率以及最佳稠油注汽转周措施方案编制效率,提升稠油井智能决策分析管理能力,提高采油管理区的效益开发水平。Based on the production characteristics of steam huff and puff in heavy oil wells and the historical data of steam injection and production in Shengli Oilfield,a sample library is constructed according to the classification of liquid volume and water content.A long short-term memory network(LSTM)algorithm is used to establish a prediction model for the timing of heavy oil well rotation,and the model is continuously trained and optimized to predict the production indicators of heavy oil wells in the next 90 days.This achieves multi-dimensional intelligent prediction of heavy oil well steam injection rotation,improves the accuracy of heavy oil well production prediction,the accuracy of the best timing prediction for heavy oil steam injection rotation,and the efficiency of the preparation of the best measures for heavy oil steam injection rotation.It enhances the intelligent decision-making,analysis,and management capabilities of heavy oil wells,and improves the development level of oil production management areas.

关 键 词：稠油转周 转周时机预测 周期参数预测 

分 类 号：TE357[石油与天然气工程—油气田开发工程]