准确预测储层参数对地下储层表征、气藏模式构建、产能释放及流体运移理解具有关键意义。传统基于岩心测量或数学-岩石物理建模的方法受限于弹性参数反演结果的多解性和低精度，难以满足现代勘探需求。

为提升低渗储层参数预测的准确性，提出了一种岩石物理建模引导的低渗储层参数预测方法。其利用卷积神经网络（Convolutional Neural Network，CNN）作为深度学习框架，从实际地震数据中直接预测含水饱和度、泥质含量及孔隙度；为解决标签数据稀缺问题，结合岩石物理建模与弹性参数随机扰动技术，生成高质量训练样本，有效扩充了数据集。

理论模型测试表明：在储层参数对岩石物理敏感性较低的情况下，也能实现低渗储层参数的空间分布预测；相比纯数据驱动的深度学习，仅需少量测井数据即可获得高精度的储层参数预测结果。在莺歌海盆地东方区的应用实践表明，该方法优化了钻井部署，助力了低渗领域的重大勘探突破和储量发现。

Accurately predicting reservoir parameters is significant for characterizing subsurface reservoirs, establishing gas accumulation patterns, releasing production capacity, and understanding fluid migration. In practical explorations, reservoir parameters such as porosity and water saturation are typically predicted using methods based on core measurement or mathematical-petrophysical modeling. However, the conventional prediction methods mentioned above rely heavily on both the inversion results of elastic parameters and corresponding petrophysical law templates, and their results feature strong multiplicity of solutions and low accuracy.

To more effectively predict reservoir parameters and guide exploration well deployment and production capacity release, this study proposed a petrophysical modeling-guided method for predicting parameters of low-permeability reservoirs. With the convolutional neural network (CNN) as a deep learning framework, the proposed method can predict water saturation, clay content, and porosity based on actual seismic data. Additionally, considering insufficient labeled data for deep learning in practical explorations, this study developed a petrophysics-guided method for constructing labeled data. Specifically, the petrophysical modeling combined with the random perturbation of elastic parameters was adopted to produce training samples, thus effectively expanding the size of sample data.

The theoretical model tests demonstrate that the petrophysical modeling-guided method for predicting parameters of low-permeability reservoirs enjoys the following advantages: (1) This method can effectively predict the spatial distributions of parameters of low-permeability reservoirs in the case of low sensitivities of reservoir parameters to petrophysics; (2) Compared to data-driven deep learning, this method can yield high-accuracy predicted results of reservoir parameters based on merely a few log data. As substantiated by exploration in the Dongfang block of the Yinggehai Basin, the proposed method facilitates the optimization of well deployment, guiding the achievement of significant exploration breakthroughs and reserve discovery in the low-permeability areas of the basin.