径向基(RBF)神经网络法具有网络结构简单、逼近能力强和学习速度快等优点,已成为最具发展潜力的储层敏感性智能预测方法之一,但在实际应用中仍存在泛化能力不强、网络训练不收敛等问题.通过在输入层中引入补充节点,对网络拓扑结构进行优化,有效地提高了RBF神经网络的逼近精度和泛化能力.在确定储层敏感性主要影响因素的基础上,通过对径向基函数散布常数的优选,进一步优化了RBF神经网络的性能.采用所收集的胜利、辽河、大港及江苏油田共125组数据,进行了神经网络训练和预测检验,优化了RBF神经网络,并在储层敏感性预测方面进行了应用.结果表明,对于训练集内的样本,预测的平均准确率均大于93.79％,且预测值与实验值的相关系数均大于0.995；对于训练集外的样本,预测的平均准确率大于91.59％,预测值与实验值的相关系数大于0.994,实现了对储层敏感性的准确、定量预测.%RBF (Radial Basic Function) neural network has many advantages,such as simple network structure,strong approaching capability, quick convergence, and so on, which is one of the most promising methods for predicting reservoir sensitivity at present. However, there are still several problems such as bad training convergence and generalization in practical application. By means of the addition of an adequate input node,radial basis function neural network topologies have been optimized. On the basis of determining the reservoir sensitivity influencing factors, via the radial substratum dispersion constant optimization, the network performance is further improved. Adopting the collected 125 groups data from Shengli, Liaohe, Dagang and Jiangsu oilfield, the RBF neural network has been trained and tested,and the application method of the modified RBF neural network in reservoir sensitivity prediction is established. Results show that,the new RBF neural network has better applicability. For the samples in the training set,quantitative prediction average accuracy rate of sensitivity index by this method is higher than 93. 79% ,especially the correlation coefficient between predicted value and experiment values is more than 0.995. For the samples outside the training set, quantitative prediction average accuracy rate of sensitivity index by this method is also above 91. 59% , and the correlation coefficient between predicted value and experiment values is more than 0.994 as well. The new method has effectively realized the goal of accurate and quantitative prediction for reservoir sensitivity.
展开▼
