【摘要】 储层孔隙度是表征储油物性、建立各类地质模型的重要参数。支持向量回归机(SVR)凭借良好的非线性回归能力,在孔隙度预测中开始广泛应用。由于不同岩性的储层孔隙类型不同,孔隙度结构也存在较大差异,导致该方法的实际应用效果仍不理想。针对上述问题,在孔隙度预测模型中考虑了岩性信息,将样本岩性转化为一种与岩性变化相关性好的属性值,以此构造一种新的预测模型。使用网格粗选和网格精选相结合的方法,优选模型参数。网格粗选确定最优解的近似范围,网格精选可以在局部区间搜索到最优解。结果表明:利用优选参数建立的预测模型,在实际资料预测结果中,加入岩性信息可以提高储层孔隙度的预测精度,该方法可行。更多还原

【Abstract】 This paper introduces a novel prediction model building on the deeper understanding that reservoir porosity is an important parameter by which to represent reservoir characteristics and thereby establish diverse geological models. This model is an alternative to SVR which has come into a wider use in the porosity prediction thanks to its ascendant nonlinear regression capability,but has been thwarted by the occurrence of the different types of reservoir pores with different lithology and more variations found in pore structure of the reservoirs. The model is developed by considering the lithology information and transforming the lithology information of the sample into a kind of attribution with a better relativity with lithology change by using the information confusion method. The model parameters are optimized by combining rough screening which determines the approximate scope of the optima with fine screening which provides the optima in a certain range. The results demonstrate that the model built with preferred parameters features a better precision in the practical application after provided with the lithology information.更多还原