一种MPSO-BP-RBF网络模型及其在石油储层预测中的应用  

An MPSO-BP-RBF Neural Network Model and its Application in Oil Reservoir Forecasting

作　　者：於世为[1,2] 诸克军[1] 郭海湘[1] 

机构地区：[1]中国地质大学经济管理学院,武汉430074 [2]北京理工大学能源与环境政策研究中心,北京100081

出　　处：《系统管理学报》2011年第2期129-135,共7页Journal of Systems & Management

基　　金：国家自然科学基金资助项目(70573101);教育部博士点科研基金(20100145120008);中国地质大学(武汉)中央高校基本科研业务费专项资金资助项目(CUG090113)

摘　　要：为适应复杂油气储集层非均匀性、非线性及不确定性的响应特征,提高储层预测精度,采用PSO混合编码,提出了一种基于混合MPSO-BP的RBF自构建学习模型。该模型中,每个粒子由整数与实数两部分构成,分别对RBF的基函数的个数及相关参数(中心,宽度,输出层权值)进行编码。同时,设计了一个特殊的适应度函数,在保证精度的前提下,使网络的结构相对简单。应用于储层预测实践中,相对于RBF其他学习算法,该算法隐节点少,精度高,泛化能力强。In order to accommodate non-uniformity, non-linearity, and uncertainty and improve forecast accuracy of a complex oil and gas reservoir, this paper proposes a hybrid self-generated training algorithm of RBF neural networks, which combines multi-encoding particle swarm optimization and back propagation （MPSO-BP） algorithms. In the algorithm, every particle consists of binary and real parts, which corre- spond to the coding of the number of hidden units and associated parameters （i. e. , widths, centers and weights）, respectively. Furthermore, a special fitness function is introduced to ensure accuracy of the al- gorithm and to improve self-adaptability and generality of the RBF neural networks. Applications in oil reservoir prediction show that the proposed method has fewer hidden nodes, higher accuracy, and better generality compared with other methods.

关 键 词：混合编码 粒子群算法 自构建 RBF网络 储层预测 

分 类 号：F830.49[经济管理—金融学]