基于BP-GA算法的水平井智能压裂设计方法  

BP-GA algorithm assisted intelligent horizontal well fracturing design

作　　者：宋丽阳[1] 王纪伟 刘长印[1] SONG Liyang;WANG Jiwei;LIU Changyin(Petroleum Exploration and Production Research Institute,SINOPEC,Beijing 100083,China)

机构地区：[1]中国石化石油勘探开发研究院,北京100083

出　　处：《断块油气田》2022年第3期417-421,426,共6页Fault-Block Oil & Gas Field

基　　金：中石化科研项目“大牛地山西组未动用储量区高效压裂关键技术”(P19019-7)。

摘　　要：针对常规水平井压裂正交方案优化设计模拟量大、历史拟合难以准确反演现场实际的问题,采用BP神经网络与遗传算法建立产能预测模型,形成水平井压裂遗传式优化设计方法。将现场数据导入BP神经网络系统,基于正反向训练与误差识别校正演算参数,预测水平井压后产能,与数值模拟、现场试验数据进行对比,评估模型精度;采用遗传算法建立压裂方案变异进化机制,基于种群遗传进化生成最优方案。研究表明:对于压裂综合甜点系数平均值高且横向差异度低的储层,采用多段少簇密集式压裂方案可显著提高压裂井产能;对于压裂综合甜点系数横向差异度高的储层,采用高甜点多簇、一段一策式精准压裂方案可最大程度提高压裂改造效果;相对于等间距固定参数布簇,采用不等间距各段差异式布缝方案,可降低缝间干扰,促进裂缝充分扩展,提高有效改造体积和裂缝系统整体导流能力。The conventional horizontal well fracturing orthogonal optimization design requires a vast number of simulation,and the history matching method can not accurately predict the field situation.In this paper,BP neural network and genetic algorithm methods are employed to develop productivity prediction model and form a genetic optimization design method for horizontal well fracturing.The field data are introduced into BP neural network system to predict horizontal well productivity based on forward and reverse training and calculation parameter of error correction,and the model accuracy is judged based on the comparation with the field test and numerical simulation results.Genetic algorithm was used to establish the mutation evolution mechanism of fracturing scheme,and the optimal scheme was generated based on population genetic evolution.The results show that for the reservoir with high average comprehensive fracturing sweet coefficient value and low transverse heterogeneity,the high-stage-number with low-cluster-number per stage fracturing scheme can significantly improve the fracturing well productivity.For the reservoir with low average comprehensive fracturing sweet coefficient value and high transverse heterogeneity,the accurate fracturing design with high-cluster-number in high sweet coefficient and one design per stage can significantly improve the fracturing effect.Compared with the uniform spacing and fixed-parameter scheme,differential fracturing design with the non-uniform spacing leads to lower fracture interaction and more fully expansion of fractures,and improve the stimulated reservoir volume and the overall conductivity of the fracture system.

关 键 词：机器学习 BP神经网络 遗传算法 水平井 压裂设计 

分 类 号：TE357.11[石油与天然气工程—油气田开发工程]