【摘要】 多元线性回归与神经网络的组合已被广泛应用于多属性储层预测,但在神经网络训练时,对已知井数目总是存在很大的依赖性,而在勘探初期往往钻井数据又很少。为解决这一问题,提出可以从现有的地震属性数据中,挖掘出更多的有效信息来弥补测井参数的缺乏。即首先利用模糊自组织神经网络将一组地震属性进行聚类映射,反映出不同沉积环境下不同地震响应的特征区域,即地震相;然后结合地质和开发资料,对地震相进行解释,并用地震相的信息约束储层预测。地震相约束可以从两方面进行,一是增加新的、合理的虚拟井参数,二是将地震聚类数据作为新的输入属性。按此思路对模糊自组织聚类、多元线性回归和径向基函数网络进行了组合,通过数学推导证明了该流程的可行性。实际资料处理、解释表明,无论时移监测中的压力差预测,还是勘探初期储层的含油饱和度预测,该流程都能获得理想的效果。更多还原

【Abstract】 Integrated multiple linear regression and radial basis function neural network is an effective quantitative approach in reservoir prediction with seismic multiattribute,but the effect of neural network depend on number of the wells used.Generally,a few wells have been drilled in the preliminary stage of exploration,and the lack of training samples leads to failure of the pattern recognition.To solve this problem,more geological information should be found to make up for the lack of well parameters based on attributes.Fuzzy self-organizing neural network is able to cluster a set of seismic attributes.The mapping reflects the different seismic response of sedimentary environments under the different regions,which is called seismic facies.Combination of geology and development data,seismic facies was interpreted.Then,information of seismic facies as constraint was added in reservoir prediction by two ways.One is that reasonable imaginary wells parameters should be added,and the second way is that clustering data will be an extra input attribute.in this way,the paper integrates fuzzy self-organizing cluster,multiple linear regression and radial basis function.The process is feasible by mathematical deducing.Meanwhile,in practical applications such as the pressure difference monitoring and the oil saturation estimation,the approach works well.更多还原