神经网络反演在火山岩储层预测中的应用

川西南井研地区二叠系火山岩分布稳定，地震反射特征整体呈“两强波峰夹一波谷”,火山岩内部发育两套储层，且上部储层由厚度小于7 m的多个薄层组成，受地震分辨能力限制，常规地震属性和波阻抗反演等方法均不能精确识别两套储层及其分布。利用孔隙度可表征储层特征且与波阻抗具有良好拟合关系的特点，提出采用拓频地震数据和神经网络孔隙度非线性反演技术预测火山岩上、下储层。地震数据经谱分解处理后得到低、中、高频数据体及其对应的分频地震属性，通过神经网络反演建立分频地震属性与孔隙度的非线性映射关系，进而得到高分辨率的孔隙度反演结果。综合预测结果表明：神经网络反演结果与实钻井吻合；纵、横向分辨率明显提高，能有效识别井研地区火山岩两套储层在纵向上主要发育在中、下部，平面上主要发育在工区西部。研究结果可指导后期勘探评价或开发井部署。

Application of neural network inversion in prediction of volcanic rock reservoir

The stratigraphic distribution of Permian volcanic rocks in the Jingyan area of southwest Sichuan is stable,and the seismic reflection is characterized by "two peaks and one trough" between them overall.There are two sets of reservoirs developing in the volcanic rocks, and the upper reservoir is composed of multiple layers with thicknesses less than 7 m.Limited by conventional seismic resolution,seismic attribute and wave impedance inversion cannot effectively identify the distribution of the two sets of reservoirs.However,porosity can represent reservoir characteristics and has a good fitting relationship with wave impedance.Therefore,seismic data with extended frequency and the nonlinear inversion of porosity with a neural network are used to predict the upper and lower volcanic reservoirs.With the spectral decomposition technology, the seismic data is transformed into low-,medium-,and high-frequency data volumes,and the corresponding frequency-divided seismic attributes are obtained.On this basis,neural network inversion is performed to establish the nonlinear relationship between porosity and frequency-divided seismic attributes.Finally,high-resolution porosity inversion results are obtained.The prediction results show obvious improvement in both vertical and horizontal resolution. The results of neural network inversion are consistent with the actual situation of drilled wells:Two sets of volcanic reservoirs mainly develop in the middle and lower part vertically and are distributed in the west of the study area horizontally.The research results can guide later exploration evaluation or development well deployment.