基于径向基函数(RBF)神经网络的储层损害诊断技术研究  

Research on the Formation Damage Diagnosis Based on Radial Basis Functions Neural Network

作　　者：黄春[1] 蒋官澄[2] 纪朝凤[3] 胡冬亮[1] 申延晴 宋友贵[3] 

机构地区：[1]中石化胜利油田,山东东营257000 [2]中国石油大学石油天然气工程学院,北京102249 [3]中石油大港油田分公司采油院,天津300280 [4]斯伦贝谢公司,天津300280

出　　处：《应用基础与工程科学学报》2010年第2期313-320,共8页Journal of Basic Science and Engineering

基　　金：国家科技重大专项(2008ZX05009-05)

摘　　要：在石油勘探开发过程中会因各种原因造成储层损害,降低油井产量甚至停产,必须对损害储层进行准确定量诊断的基础上采取相应的解堵措施,提高或恢复油井产量.人工神经网络法是进行储层损害诊断较好的方法,但目前仅限于BP神经网络,或改进的BP神经网络的应用.本文通过对BP神经网络和径向基函数(RBF)网络的对比表明,径向基函数(RBF)网络具有收敛速度快、预测精度高等优点,并在确定影响储层敏感性和各损害类型因素的基础上,分别收集了各数据70组以上,然后进行了径向基(RBF)网络训练和应用,分别建立了径向基(RBF)神经网络在储层损害敏感性和定量诊断领域的应用,实现了对储层水敏性、盐敏性、速敏性预测,以及储层微粒运移、固相颗粒堵塞、水化膨胀和结垢损害程度的定量诊断;同时,提出的径向基(RBF)网络储层损害定量诊断法在孤东油田17口井上得到了检验,成功率达100%,证明RBF网络法与其它方法相比具有诊断结果准确性高、结论可靠、推广应用方便、收敛速度快等优点,提高了储层保护措施和解堵措施优化决策的科学性和准确性,在油田生产中发挥了重要作用.In the production course of oil exploration, formation damage due to various reasons will cause oil well decrease even stop production. Based on accurately quantitative diagnosis of formation damage, plugging removal must be adopted accordingly to improve or restore oil well production. Artificial neural network is a better method for diagnosis on formation damage, but it is restricted to the BP network at present or improved application of BP neural network. In the paper, the BP neural network and radial basis function （RBF） network was compared, which showed that RBF network had some advantages such as higher convergence rate, higher prediction prevision and so on. After determining the influencing factors of damage types and formation sensitivity, over 70 sets of data about each factor were collected. Then training and application on RBF network was performed and the application of RBF network in the area of formation damage sensitivity, and quantitative diagnosis was established respectively, which realized the quantitative prediction of water sensitivity, salt sensitivity and velocity sensitivity and the diagnosis of damage degree on migration of formation solid particles, solid plugging, hydration swelling and scaling. Meanwhile, the RBF network quantitative diagnosis method was tested in 17 wells of Gudong oilfield and it had a success rate of 100%. It proved that, compared with other methods, RBF network has some advantages, for example, the result of diagnosis is more accurate, the conclusion is more reliable, the popularization and application is more convenient and the rate of convergence is faster. Therefore, RBF network method improves the accuracy and the scientificity of optimization decision for the measures of formation protection and plugging so that it plays an important role in oilfield production.

关 键 词：储层损害 诊断 径向基网络 BP神经网络 人工神经网络 

分 类 号：TE258[石油与天然气工程—油气井工程]