基于混合降维Elman神经网络的砂岩储层物性参数智能计算研究  

Research of Sandstone Reservoir Physical Properties Estimation Based on Elman Neural Networks with Hybrid Dimensionality Reduction

作　　者：程国建[1] 马微[1] 刘烨[1] 魏新善[2] 荣春龙[2] 南珺祥[2] 

机构地区：[1]西安石油大学计算机学院,西安710065 [2]中国石油长庆油田分公司勘探开发研究院,西安710021

出　　处：《科学技术与工程》2014年第3期24-28,32,共6页Science Technology and Engineering

基　　金：国家自然科学基金项目(40872087)资助

摘　　要：针对实验室测定岩石储层物性参数在实际应用中的成本问题,提出一个基于混合降维Elman神经网络的砂岩储层物性参数计算的智能方法。首先利用灰色关联分析对岩石薄片特征参数与其物性参数进行关联度计算,优选关联度较高的若干参数;其次使用主成分分析对选出的特征参数二次降维,最后应用Elman神经网络寻找岩石薄片特征参数与其物性参数之间的映射关系。选取鄂尔多斯盆地吴旗地区薛岔区块延长组储层砂岩样本的薄片鉴定与物性分析数据对方法进行测试,实验结果表明,计算得到的孔隙度与渗透率平均相对误差分别为7.28%和6.25%,混合降维方法在收敛速度和计算精度方面也得到提高。因此,基于混合降维Elman神经网络方法能够利用成本较低的岩石薄片相关资料快速并准确地计算砂岩储层物性参数,具有较高的可靠性、实用性以及应用前景。For the high economic cost of rock physical properties tested in laboratory, an intelligent method which based on Elman neural networks with the hybrid dimensionality reduction, was proposed to estimate the prop- erties of sandstone. With grey correlation analysis method, the correlation between rock slice characteristic parame- ters and physical properties were built. Then the appropriate parameters with high correlation would be chosen, and the second dimensionality reduction process was utilized for these parameters by principal component analysis meth- od. Finally, the mapping relationship between rock slice characteristic parameters and physical properties had been found through Elman neural networks. The estimation validity and reliability for this method were tested with practi- cal data from Xuecha, Wuqi region in Ordos Basin. The result showed that the average relative errors of porosity and permeability estimation for this method were 7.28% and 6. 25% respectively, and this method had the better convergence speed and more accuracy than others. Therefore, by using the cheap rock slice related information, the rock reservoir physical parameters can be estimated efficiently and accurately, and it is of high reliability, prac- ticability and application prospect.

关 键 词：混合降维 ELMAN神经网络 砂岩储层 物性参数 智能计算 

分 类 号：TE155[石油与天然气工程—油气勘探] TP183[自动化与计算机技术—控制理论与控制工程]