论文详情
裂缝发育程度会影响地震波动力学特征和地震波同相轴的形态,但岩性、物性、流体性质的改变也会影响上述特征的变化。因此,利用单属性预测裂缝会有多解性,地震多属性综合预测裂缝是减少多解性的有效措施。对于多特征输入的预测问题,机器学习有其独特的优势,其中具有较强泛化能力和运算效率的极限学习机算法值得重点考虑。为此,在裂缝发育带预测中引入了极限学习机算法。首先基于测井数据,利用极限学习机预测裂缝发育状况并将预测结果与近似支持向量机分类效果进行对比;然后,利用井旁道地震属性数据进行裂缝识别,分析极限学习机在裂缝预测中的效果与优势;最后通过极限学习机算法对地震属性特征与裂缝带发育程度之间对应关系的学习,将其应用于实际工区。结果表明,相较于近似支持向量机,极限学习机在保证分类准确度的同时训练效率更高,能够综合多种地震属性刻画大尺度裂缝带,实现致密砂岩裂缝储层裂缝带发育程度的有效预测,为裂缝的综合预测提供了新的思路。
The degree of fracture development,changes in lithology,petrophysical properties,and fluid properties will affect the dynamic characteristics of seismic waves and the shape of seismic wave events.If fractures are predicted on the basis of a single attribute,multiple solutions are found,and the problem remains undetermined.Carrying out a comprehensive,multi-attribute seismic prediction can be an effective way to reduce the number of solutions.Extreme learning machines (ELMs) possess unique advantages when solving prediction problems with multi-attribute inputs owing to their strong generalization capabilities and computational efficiency.In this study,an ELM was utilized for the prediction of a fracture zone from well logging data.The ELM’s output was compared with that of a proximal support vector machine.Subsequently,the effect of the ELM on fracture prediction was investigated using the near-well seismic attribute.Finally,the ELM algorithm was used to identify the relationship between seismic attribute characteristics and fracture zone development,which was then applied to field data.The results showed that compared with the proximal support vector machine,the ELM offers better training efficiency while ensuring classification accuracy.It can be used to characterize large-scale fracture zones based on a variety of seismic attributes and can effectively predict fracture zones in tight sandstone fracture reservoirs,providing valuable support in the comprehensive prediction of fractures.
国家自然科学基金“基于频变信息的流体识别及流体可动性预测”(41774142)和国家科技重大专项“致密砂岩储层有效裂缝预测方法研究”(2016ZX05002-004-013)共同资助。