论文详情
沉积微相划分是油气藏勘探的研究基础。传统沉积微相划分由地质学家根据自身掌握的知识和经验手工完成,这种地质学家主导的人工解释是主观的、冗长的,可能引入人为偏差。深度学习在解决复杂非线性问题上具有优势,目前尚无有效解决沉积微相划分的深度学习方法。针对测井沉积微相,提出了基于特征构造(DMC)和双向长短期记忆网络(BiLSTM)的沉积微相智能识别方法。首先,利用趋势分解和中值滤波对原始曲线进行多维重构,并对重构矩阵和原始曲线特征采用kmeans提取时空相关聚类特征;然后,以原始曲线特征、地质趋势特征、中值滤波特征和聚类特征作为输入,基于双向长短期记忆网络得到当前深度沉积微相预测类型。与长短期记忆网络(LSTM)和时间卷积网络(TCN)对比发现,在沉积微相的识别上,沉积微相智能识别方法具有更优异的性能和鲁棒性。实验表明,该方法能有效划分沉积微相,识别准确率达到91.69%。
Division of sedimentary microfacies is the basis of oil and gas exploration.This is usually performed manually by geologists based on their knowledge and experience.However,manual identification is time-consuming and subjective,and can lead to human bias.Deep-learning algorithms can solve complex nonlinear problems but they have not been applied to sedimentary microfacies identification thus far.In this study,a method for the intelligent identification of sedimentary microfacies based on feature structure (DMC) and a bidirectional long short-term memory network (BILSTM) is proposed.First,the original curve was reconstructed by trend decomposition and median filtering,and the spatial and temporal correlation clustering features were extracted by the k-means method from the reconstructed matrix and original curve features.Then,by inputting the characteristics of the original curve as well as those of the geological trend,median filtering,and clustering,the sedimentary microfacies at a given depth were predicted based on BILSTM.Compared with the long-short term memory network (LSTM) and time convolution network (TCN),the proposed method performs better and is more robust.Experimental results showed that sedimentary microfacies can be classified by the proposed method with a recognition accuracy of 91.69%.
国家重点研发计划深地专项项目(2016YFC0601100)和四川省科技计划项目(2019CXRC0027)共同资助。