基于PSO-ELM的水平井自喷期“多段式”产量预测方法——以玛湖油田百口泉组致密砾岩油藏为例  

PSO-ELM Based“Multi-stage”Production Prediction Method for Horizontal Wells with Self-injection Period:A Case Study of Dense Conglomerate Reservoir in Baikouquan Formation of Mahu Oilfield

作　　者：王林生[1] 黄长兵 朱键[3] 覃建华[3] 张景[3] 李文涛 WANG Lin-sheng;HUANG Chang-bing;ZHU Jian;QIN Jian-hua;ZHANG Jing;LI Wen-tao(Petro China Xinjiang Oilfield Company,Karamay 834000,China;Key Laboratory of Solid Waste Treatment and Resource Recycle,Ministry of Education,Mianyang 621010,China;Institute of Exploration and Development,Xinjiang Oilfield Company,Karamay 834000,China)

机构地区：[1]中石油新疆油田公司,克拉玛依834000 [2]西南科技大学固体废物处理与资源化教育部重点实验室,绵阳621010 [3]新疆油田公司勘探开发研究院,克拉玛依834000

出　　处：《科学技术与工程》2023年第5期1931-1936,共6页Science Technology and Engineering

基　　金：国家科技重大专项(2017ZX05070);中国石油重大科技专项(2017E-04)。

摘　　要：准确预测油气井动态产量对油田高效开发意义重大,是单井累产油预测以及部署政策优化的关键。玛瑚油田百口泉组致密砾岩油藏水平井自喷期产量呈“多段式”特征,在实际生产过程中,油气井产量受储层物性、压裂工艺参数等多种因素综合影响,传统产量预测方法及数值模拟法考虑影响因素有限,预测方法适用性差。在产量特征认识基础之上,利用主成分分析法(principal component analysis,PCA)优选油层厚度、地层压力、总砂量、渗透率、压裂簇数及含油饱和度6个主控因素,采用粒子群优化-极限学习机(particle swarm optimization-extreme learning machine,PSO-ELM)的输入权值与隐含层偏置,建立了玛湖油田水平井产量预测模型。预测结果表明,PSO-ELM对比传统预测模型具有计算速度快、泛化能力强、预测精度高的优点,利用该方法预测了5口水平井的单井产量,平均误差在2.14%~5.28%,与实际产量吻合良好。Accurate prediction of dynamic production of oil and gas wells is of great significance to the efficient development of oil fields,and is the key to the prediction of cumulative oil production of single wells and optimization of deployment policies.In the actual production process,the production of oil and gas wells is influenced by various factors such as reservoir properties and fracturing process parameters,etc.Traditional production prediction methods and numerical simulation methods have limited consideration of the influencing factors and poor applicability of the prediction methods.Based on the understanding of production characteristics,the six main control factors including formation thickness,formation pressure,total sand volume,permeability,number of fracture clusters and oil saturation were selected by using principal component analysis(PCA),and the input weights and hidden layer bias were optimized by using particle swarm optimization-extreme learning machine(PSO-ELM)to establish a horizontal well production prediction model in Mahu oilfield.The prediction results show that PSO-ELM has the advantages of fast calculation speed,strong generalization ability and high prediction accuracy compared with the traditional prediction model,and the single well production of five horizontal wells was predicted by the method with an average error of 2.14%~5.28%,which is in good agreement with the actual production.

关 键 词：多段式 产量预测 主成分分析 粒子群优化 极限学习机 

分 类 号：TE328[石油与天然气工程—油气田开发工程]