一种模糊神经网络技术及其在储层预测中的应用

Fuzzy neural network and its application in reservoir prediction

大庆油田有限责任公司勘探开发研究院,黑龙江大庆 163712

Exploration and Development Research Institute of Daqing Oilfield Company Limited, Daqing 163712, China

大庆外围油田的葡萄花油层主要为砂泥岩薄互层,储层砂体横向变化大,这给井位设计带来了很大的难度。近几年地震属性分析技术虽然得到了较快的发展,但地震属性与储层地质参数之间的关系较模糊,难以用地震特征参数直接预测储层的砂岩厚度。为此,研究了一种模糊神经网络预测砂岩技术,它将人工神经网络理论与模糊逻辑分析相结合,在地震属性分析的基础上,以井旁地震道主分量参数为输入,以井孔地质参数为期望输出,建立模糊神经网络,并对网络进行训练,当网络收敛且网络整体方差达到要求的精度时,便完成了网络训练。该技术应用于大庆太平屯地区储层预测中,通过4口后验井检验,预测厚度与钻井厚度吻合较好,平均绝对误差为0.21m。

It is hard to estimate the sandstone thickness in reservoir intervals because there is lack of direct relationship between seismic attributes and reservoir parameters. This paper presents a fuzzy neural network technique for sandstone prediction. The technique combines artificial neural network theory with fuzzy logic analysis. Based on seismic attribute analysis, a fuzzy neural network was established which used the principal components of traces nearby wells as input and wellbore geologic parameters as output. The network was then be trained until it converged such that its overall variance was less than the predefined precision. This technique has been applied in the reservoir prediction in Taipingtun region of Daqing. Verifications in four wells showed that the average absolute thickness error is 0.21m.