基于粒子群优化算法的小波神经网络缝洞型储层识别模型  

THE FRACTURED-VUGGY RESERVOIR IDENTIFICATION MODEL OF PSO-BASED WAVELET NEURAL NETWORK

作　　者：罗鑫[1] 匡建超[1] 瞿子易[1] 曾剑毅[1] 

机构地区：[1]成都理工大学能源学院,四川成都610059

出　　处：《物探化探计算技术》2009年第3期213-217,174-175,共5页Computing Techniques For Geophysical and Geochemical Exploration

基　　金：四川省学术与技术带头人培养资金(川人办发[2008]24号)

摘　　要：针对缝洞型储层识别精度较低这一难题,提出了基于粒子群优化算法的小波神经网络(PSO-WNN)储层识别模型。以小波函数作为隐含层的激励函数,采用粒子群优化算法,对权值、伸缩参数、平移参数进行调整,构建出基于粒子群优化算法的小波神经网络储层识别模型。该模型具有算法简单、结构稳定、计算收敛速度快、全局寻优能力强、识别精度高、泛化能力强的优点。这里以济阳坳陷桩西埕岛地区古生界潜山缝洞型储层识别为例,利用常规测井参数作为模型的输入参数,以储层类型赋值作为输出,选取九口井的108个已知样本,采用不同隐含层个数对模型进行多次训练。通过对比分析,最终确定隐含层个数为10,建立起该区的Ⅰ类、Ⅱ类、Ⅲ类储层识别模型。利用已建模型对十八个检验样本进行识别,其识别正确率高达100%,而BP神经网络识别正确率为88%。这表明该模型对缝洞型储层的识别效果较好,为缝洞型储层的进一步研究提供了可靠的依据。It is difficult to accurately identify reservoir, this paper proposes a model of wavelet neural networks based on PSO (PSO-WNN) to solve the problem.The activation function which is in the hidden layer is wavelet function.The fracture-caves reservoir identification model is formed based on PSO-based Wavelet Neural Network, and PSO is used to adjust the parameters of weight, stretching, parallel move.The model has the advantage of simple algorithm, structural stability,fast speed of convergence,global optimization ability,high-accuracy of identification,generalization ability,etc.108 samples of 9 testing wells from the reservoir of buried-hill in Zhuangxi-Chengdao,Jiyang sag were selected, and the input parameters are conventional logging parameters and the output parameter is the assignment of the type of reservoir.In the identification,different number of hidden layer has been used to train the model and the final number of hidden layer is finally decided as 10 through comparative analysis.The model of Ⅰ,Ⅱ, Ⅲ type of reservoir is established. In the 18 samples for check, the correction rate is 100% while BP is 88%,which shows that the model is better for fracture-caves reservoir. The method is providing a reliable basis for the further study of fracture-caves reservoir.

关 键 词：缝洞型储层 粒子群优化算法 小波神经网络 储层识别 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]