基于长短期记忆网络的钻前测井曲线预测方法

Method of well logging prediction prior to well drilling based on long short-term memory recurrent neural network

基于深度学习的最新成果,提出了一种基于长短期记忆(long short-term memory)循环神经网络的钻前测井曲线预测方法,使用该方法能从已钻地层段及邻域内获得的测井数据预测钻前的测井曲线,进而获得钻前的地层岩石信息,解决油气钻探过程中测井曲线只能在钻后获得的滞后性,以提高钻前地层构造及压力预测的准确性。将其与普通循环神经网络的预测结果进行对比分析,结果表明,长短时记忆网络建模预测效果良好,能比较准确地预测钻前测井曲线的变化趋势,是一种有效且预测精度较高的钻前测井曲线预测方法。

According to the latest achievement of deep learning,apre-drilling logging data prediction method based on long short-term memory(LSTM)recurrent neural network is proposed through the accurate improvement of the pre-drilling stratum structure and pressure prediction,in order to solve the hysteresis problem that the logging data can only be obtained after drilling in the oil and gas drilling process.This method can be used to predict the pre-drilling logging data from the logging data obtained in the drilled strata and adjacent areas,and then to predict the pre-drilling rock information.Compared with the prediction results of the general recurrent neural network,the proposed experimental results are showed that the long-short time memory network has a good prediction effect and can accurately predict the change trend of the pre-drilling logging data.Therefore,the new predrilling logging data prediction method based on long short-term memory(LSTM)recurrent neural network is an effective and accurate method in the pre-drilling well logging prediction.