径向基神经网络油气层损害诊断方法  

Method of formation damage diagnosis based on radial basis neural network

作　　者：李师涛[1] 蒋官澄[2] 陈应淋[1] 

机构地区：[1]中国石化股份胜利油田分公司东胜精攻石油开发集团股份有限公司,山东东营257000 [2]中国石油大学(北京)石油天然气工程学院,北京102249

出　　处：《油气地质与采收率》2009年第6期98-101,共4页Petroleum Geology and Recovery Efficiency

基　　金：国家重大油气专项"复杂结构井油气层保护技术研究"(20092X05009-005)

摘　　要：在诊断油气层损害方面,人工神经网络具有许多优越性,尤其是BP神经网络,但BP神经网络存在的一些缺点限制了它的推广应用。通过对BP神经网络和径向基神经网络的对比表明,径向基神经网络具有收敛速度快和预测精度高等优点,其网络模型的预测绝对误差平均为13.89%,而L-M优化算法网络的为32.63%。建立了径向基神经网络在油气层损害诊断领域的应用方法,对油气层敏感性和损害程度进行了预测,网络预测值和实际值的相关系数达0.991以上,准确率大于80%。该方法在孤东油田得到了很好的应用,成功率达100%,实现了对油气层损害类型和程度的定量诊断,与其他方法相比具有诊断结果准确性高、推广应用方便、收敛速度快等优点。Artificial neural network has many advantages of formation damage diagnosis,especially the BP neural network,but its application is limited for some shortcomings. Comparison between BP neural network and radial basis neural networks shows that the latter has the advantages of fast convergence and high prediction precision,whose network model forecasts the average absolute error only 13.89% ,while the LM optimization algorithm network of 32.63%. Radial basis neural network was applied in the formation damage diagnosis. Sensitivity and damage degree to the reservoir were predicted ,with the correlation coefficient between the network predicted value and actual value above 0.991 and the accuracy greater than 80%. The application of this method in Gudong Oilfield has gained good effort,with the success rate of 100%. The type and extent of the formation damage were diagnosed quantitative. This method has the advantages of high accuracy,convenient application and fast convergence than other methods.

关 键 词：径向基神经网络 BP神经网络 油气层损害诊断 网络预测 孤东油田 

分 类 号：TE319[石油与天然气工程—油气田开发工程]