基于LSTM堆叠残差网络的岩相识别方法

Lithofacies Identification Method Based on LSTM Stacked Residual Network

为了提高岩相识别的准确性,本文开发一种基于残差连接长短期记忆网络的非均质储层岩相智能识别模型(LSTM_res)。首先,基于长短期记忆神经网络构建序列特征模块获取测井关键特征,该模块的多层叠加进一步增强了模型对关键特征信息的提取能力;其次,在序列特征模块的基础上引入残差连接技术,实现模型对网络不同层间特征信息的提取和融合,有效解决深度神经网络的退化问题;最后,以挪威附近北海浅海地区的测井数据为研究对象,通过测井参数敏感性分析选取6种测井参数(RMED、RHOB、GR、NPHI、PEF和SP)实现储层岩相智能识别。实验结果表明,在同等条件下与LSTM、CNN_res和CNN模型相比,LSTM_res模型的岩相识别精度分别提高了2、4和6个百分点,为储层建模和地质研究提供了快速有效的数据支撑。

In order to improve the accuracy of lithofacies identification,this paper developed a heterogeneous reservoir lithofa‐cies intelligent identification model based on residual connection long short-term memory network(LSTM_res).Firstly,a se‐quence feature module is constructed based on long short-term memory neural network to obtain key logging features.The multi-layer stacking of this module further enhances the model’s ability to extract key feature information.Secondly,the residual con‐nection technology is introduced on the basis of the sequence feature module to realize the extraction and fusion of the feature in‐formation between different layers of the network,which can effectively solve the degradation problem of the deep neural net‐work.Finally,taking the logging data in the shallow sea area of the North Sea near Norway as the research object,six logging pa‐rameters(RMED,RHOB,GR,NPHI,PEF and SP)are selected through sensitivity analysis of logging parameters to realize in‐telligent identification of reservoir lithofacies.Compared with LSTM,CNN_res and CNN models under the same conditions,the experimental results show that the lithofacies identification accuracy of LSTM_res model is improved by 2,4 and 6 porcentage points,respectively.It provides fast and effective data support for reservoir modeling and geological research.