基于注意力机制的碳酸盐岩储层岩相识别方法

Lithofacies Identification Method of Carbonate Reservoir Based on Attention Mechanism

岩相识别是储层评价和油藏描述等的基础环节,碳酸盐岩储层具有非均质性强、孔隙结构复杂等特点,给岩相识别带来了挑战。融合卷积神经网络(CNN)和注意力机制开发了一种新型网络框架,根据各种测井数据之间的相关性实现储层岩相的自动识别。该框架包括特征注意力(FAtt)模块和CNN模块,FAtt模块根据识别目标与各种测井数据之间的相关性自动提取关键特征,CNN模块捕获各个测井序列之间的空间信息,两者结合有效提高了模型的岩相识别精度。基于碳酸盐岩非均质储层的实验表明,相比于单一的CNN模型,提出的模型岩相识别精度提高了9%。该模型为储层测井评价提供了一种经济可靠的替代方案,为地质研究与人工智能结合提供了快速有效的岩相数据。

Lithofacies identification is the basic link of reservoir evaluation,reservoir description and other exploration and development.Carbonate reservoir has the characteristics of strong heterogeneity and complex pore structure,which brings challenges to lithofacies identification.In this paper,a new network framework based on convolutional neural network(CNN)and attention mechanism is developed to realize automatic reservoir lithofacies identification according to the relation between logging characteristics.The framework includes feature attention(FAtt)module and CNN module.The former can automatically extract key features based on the relation between the identification targets and logging characteristics,and the latter can capture the spatial information between logging sequence.The combination of the modules effectively improves the accuracy of lithofacies identification.Compared with the single neural network,the lithofacies identification accuracy of the proposed model is improved by 9%in the experiments based on carbonate heterogeneous reservoir.This method provides an economical and reliable alternative for reservoir log evaluation,and provides fast and effective lithofacies data for the combination of geological research and artificial intelligence.