基于深度学习的水下井口弯曲应力预测方法

A Method for Predicting Bending Stress of Subsea Wellhead Based on Deep Learning

水下井口系统是海洋油气勘探开发的关键装备,为保证其在钻完井作业过程中的安全性,水下井口系统的疲劳状态监测成为保障安全的重要手段,水下井口的弯曲应力预测是基于监测数据疲劳损伤评估的重要内容。为此,提出了一种基于深度学习的水下井口系统弯曲应力预测方法,该方法使用防喷器处的加速度监测数据预测水下井口系统的弯曲应力。根据设计海况矩阵进行隔水管-水下井口系统动力响应分析,以分析得到的时序数据为基础,利用长短期记忆网络(LSTM)建立水下井口系统弯曲应力预测模型,最后对预测模型进行验证。研究结果表明,提出的水下井口系统弯曲应力预测模型得到的预测值准确度较高,相对误差低于0.2%,决定系数大于0.999。该弯曲应力预测方法可为水下井口疲劳损伤评估提供支撑。

Subsea wellhead system is a key equipment for offshore oil and gas exploration and development.To ensure its safety during drilling and completion operations,the fatigue state monitoring of the subsea wellhead system has become an important safety guarantee method,and the bending stress prediction of the subsea wellhead is an important content of fatigue damage evaluation based on monitoring data.In the paper,a bending stress prediction method based on deep learning for subsea wellhead system was proposed,which uses the acceleration monitoring data at the blowout preventer(BOP)to predict the bending stress of the subsea wellhead system.Then,the dynamic response of the riser-subsea wellhead system was analyzed under the design sea condition matrix,and the long-short term memory network(LSTM)was used to establish the bending stress prediction model of the subsea wellhead system based on the sequential data obtained from the analysis.Finally,the prediction model was verified.The results show that the prediction value obtained from the proposed bending stress prediction model of the subsea wellhead system has a high accuracy,the relative error is less than 0.2%,and the determination coefficient is greater than 0.999.The proposed bending stress prediction method provides a basis for evaluating the fatigue damage of the subsea wellhead.