基于粒子群算法的模糊优选神经网络储层识别模型  

THE RESERVOIR DISTINGUISHING MODEL OF FUZZY OPTIMUM SELECTION NEURAL NETWORK BASED ON PSO

作　　者：张银德[1] 匡建超[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》2008年第3期202-205,168-169,共4页Computing Techniques For Geophysical and Geochemical Exploration

摘　　要：针对致密储层中气水干层识别准确率较低这一难题,提出基于粒子群算法(particle swarm optimization,PSO)的模糊优选神经网络识别模型。其基本思路:首先对所有的变量进行分析,提取反映储层本质的主要属性,消除重叠信息的输入,然后将粒子群算法与模糊优选神经网络相结合,构建基于粒子群算法的模糊优选神经网络识别模型。以陕甘宁盆地中部气田马五1储层气水干层识别问题为例,选用十九口井分层测试的92个已知样本(其中八十个用于模型构建,十二个用于精度检验),对物性、测井和储渗特征等方面的十个特征参数进行分析,从中提炼出电阻率、自然伽玛、产能系数、储渗因子、介质类型因子等五个主成份控制特征参数,消除重叠信息的影响。并以此作为神经网络输入层的输入,以样本储层的产能赋值作为输出,构建基于粒子群算法的模糊优选神经网络模型。模型的识别正确率达到100%,标准误差比传统模糊神经网络降低了60%。这表明该模型具有更高的识别精度,为致密储层的准确识别探索了又一种新方法,对同类地区的研究在一定程度上具有指导作用。It is difficult to accurately distinguish gas, water and dry layer in tight reservoirs. This paper proposes a model of fuzzy optimum selection neural networks based on PSO to solve the problem. The basic thought is to find the main characters, which reflect essential of the reservoirs, by analyzing all the variables and forming models of distinguishing reservoirs using the fuzzy optimum selection neural network and PSO. By analyzing 92 samples with 10 characteristic parameters related to physical property, well logging, reservation and permeability ( with 80 samples being used to form the model and the others to examine the precision of the model) selected from the 19 testing wells in the MA51 Member in the central gas field of Shan-Gan-Ning basin, 5 main characteristic parameters of Rlld, GR, kh, KΦs and EE were obtains. The fuzzy optimum selection neural networks based on PSO was established by using the 5 characteristic parameters as input variable and making the evaluation of the productivity as output variable. The result indicated that the accuracy rate of the model is 100% and the normative error is 60% lower than the traditional fuzzy neural networks. Therefore, the model is potential for distinguishing tight reservoirs and can provide useful reference for the similar research in other regions.

关 键 词：粒子群算法 神经网络 模糊优选 储层识别 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TS210.3[自动化与计算机技术—控制科学与工程]