基于长短期记忆循环神经网络的伊拉克H油田碳酸盐岩储层渗透率测井评价

Permeability logging evaluation of carbonate reservoirs in Oilfield H of Iraq based on long short-term memory recurrent neural network

伊拉克H油田碳酸盐岩储层孔隙结构复杂,孔隙类型多样,给渗透率测井评价工作带来了极大困难。针对这一问题,建立了基于测井序列信息的长短期记忆(LSTM)循环神经网络渗透率预测模型。从测井响应差异以及测井序列信息出发,优选敏感测井曲线,搭建LSTM循环神经网络,训练网络并优化网络参数,建立了基于LSTM循环神经网络的伊拉克H油田碳酸盐岩储层渗透率预测模型。应用该模型对伊拉克H油田进行渗透率测井评价,并将预测结果与灰色系统预测模型GM(0,N)进行对比。结果表明:相对于灰色系统预测模型的结果,基于LSTM循环神经网络的渗透率预测模型的均方根误差降低了29.47%,皮尔逊(Pearson)相关系数提高了6.59%,取得了较好的应用效果。该模型能够充分挖掘测井曲线与渗透率之间关系的信息,提升了储层渗透率评价精度。

Complex pore structure and various pore types of carbonate reservoir in Oilfield H of Iraq bring much challenge to permeability logging evaluation.To solve this problem,a permeability predicting model based on long short-term memory(LSTM)recurrent neural network is established.Starting from logging response difference and logging series information,sensitive logging curves are selected,LSTM recurrent neural network is built,network is trained,network parameters are optimized and permeability predicting model of carbonate reservoir in Oilfield H of Iraq based on LSTM recurrent neural network is established.The model is applied to permeability logging evaluation of Oilfield H in Iraq,and predicted results are compared with those of grey system predicting model GM(0,N).The results show that compared with grey system predicting model,the root mean square error of permeability predicting model based on LSTM recurrent neural network is reduced by 29.47%and Pearson correlation coefficient is increased by 6.59%,with better application effect.The model can fully mine the information between logging curve and reservior permeability to increase accuracy of permeability evaluation.