基于LSTM循环神经网络的储层物性参数预测方法研究

Reservoir physical parameters prediction based on LSTM recurrent neural network

储层物性参数是反映储层油气储集能力的重要参数,表征了不同地质时期的沉积特征.地球物理测井参数由深及浅反映了不同地质时期的声、放、电等沉积特征,因而测井参数和泥质含量(孔隙度)之间有很强非线性映射关系,并具有时间序列特征.充分利用多种测井参数预测储层泥质含量和孔隙度对于储层精细描述具有十分重要的意义.深度学习技术具有极强的数据结构挖掘能力,目前,全连接的深度神经网络已经在泥质含量预测进行了初步尝试并取得了较好的效果.而长短时记忆(LSTM)循环神经网络更适合解决序列化的数据问题,因此本文提出基于LSTM循环神经网络利用多种测井参数进行泥质含量和孔隙度预测的方法,预测结果的均方根误差比常规全连接深度神经网络分别下降了42.2%和48.6%,实际应用表明,对于具有序列化特性的泥质含量和孔隙度,LSTM循环神经网络预测的准确性和稳定性要明显优于常规全连接深度神经网络.

Shale content and porosity are two important parameters for the oil and gas storage, which reflect the geological characteristics of different historical periods. The logging parameters obtained from deep to shallow strata present the stratigraphic sedimentary characteristics in different geological periods, so there is a strong nonlinear mapping relationship between shale content(porosity) and logging parameters. It is very important to make full use of logging parameters to predict reservoir’s shale content and porosity for fine reservoir description. Deep neural network technology has a strong data structure mining ability, which has been applied to predict the shale content and porosity in recent years. In fact, Long Short-Term Memory(LSTM)recurrent neural network has a further advantage in dealing with serialized data. We propose a method based on LSTM recurrent neural network to predict shale content and porosity with combinations of multi logging parameters. Compared with the traditional fully connected deep neural network at the field data, the root mean square error of prediction results decreased by 42.2% and 48.6% respectively, indicating that LSTM recurrent neural network is superior to predict the serialized reservoir parameters such as shale content and porosity.