摘要
为改善现有油气水多相流量预测时存在精度低的问题,建立了一种基于深度学习的多相流量预测模型。设计了多相流测试实验及数据采集方案,为多相流量预测提供数据支撑;设计了特征提取器、标签预测器和域鉴别器3个模块,可通过最小化损失函数实现域鉴别器区分源域和目标域特征。对油、水、空气3种不同的转移场景进行试验,并与CNN和DNN模型进行对比。结果表明,基于深度学习的多相流量预测模型性能更优,3种场景下流量预测平均绝对百分比误差(MAPE)分别为8.41%、11.05%和9.08%。仿真结果进一步验证了基于深度学习的多相流量预测模型的有效性和准确性。
In order to improve the low accuracy of the existing oil,gas and water multiphase flow prediction,a multiphase flow prediction model based on deep learning is established.Firstly,the multiphase flow test experiment and data acquisition scheme are designed to provide data support for multiphase flow prediction.Secondly,three modules,namely feature extractor,tag predictor,and domain discriminator,are designed.The domain discriminator can distinguish the source domain from the target domain by minimizing the loss function.In the experimental stage,the experimental process of three different transfer scenarios of oil,water,and air is designed,and compared with CNN and DNN models.The results show that the multiphase flow prediction model based on deep learning has better performance,and the Mean Absolute Percentage Error(MAPE)of the flow prediction under three scenarios is 8.41%,11.05%,and 9.08%,respectively.The simulation results further verify the validity and accuracy of the multiphase flow prediction model based on deep learning.
作者
李雪莹
LI Xueying(China University of Petroleum(Beijing))
出处
《油气田地面工程》
2023年第6期14-19,共6页
Oil-Gas Field Surface Engineering
