基于双向长短期记忆神经网络的岩相预测方法

介绍一种基于双向长短期记忆神经网络(Bi-directional long short-term memory,Bi-LSTM)的岩相预测方法,综合利用测井和地震数据进行高效准确的岩相预测。通过合成地震记录,进行井震数据的时深匹配,以地震吸收衰减数据、纵波阻抗、密度和伽马拟声波阻抗作为输入,以岩相作为标签,通过Bi-LSTM模型训练建立输入数据与岩相的非线性映射关系。将该方法应用于四川某浅层河道砂体勘探区岩相预测,结果表明,基于Bi-LSTM构建的岩相预测方法优于普通循环神经网络和普通LSTM,能够快速确定地下岩相,有效指示河道。基于Bi-LSTM的岩相预测方法能有效提取输入数据与岩相信息的非线性映射关系,对少井地区的岩相预测工作有较高的实用价值。

Lithofacies prediction method based on bidirectional long short memory neural network

A Bi-directional Long Short-Term Memory(BI-LSTM)based neural network method for lithofacies prediction is introduced,which combines logging and seismic data to achieve efficient and accurate lithofacies prediction.Time-depth matching of well seismic data is carried out through synthesis of seismic records.In the method,seismic absorption attenuation data,compressional wave impedance,density and gamma pseudo-acoustic impedance are taken as input,and lithofacies is taken as label.The nonlinear mapping relationship between the input data and lithofacies is established through BI-LSTM model training.The method was applied to lithofacies prediction in shallow channel sand body exploration in Sichuan area,and the results show that the method based on BI-LSTM is better than ordinary cyclic neural network and ordinary LSTM,and it can quickly determine the underground lithofacies and effectively indicate the channel.Bi-LSTM based lithofacies prediction method can effectively extract the nonlinear mapping relationship between input data and lithofacies information,which has a high practical value for lithofacies prediction in areas with few drilled wells.