融合频域注意力与趋势分解线性网络的井漏预测方法

打开文本图片集

中图分类号：TE21文献标识码：ADOI：10.12473/CPM.202410020

Abstract：Aiming at the significant periodic characteristics and irregular fluctuations oflost circulation data， and the limitations of existing methods incapturing these features，a lost circulation prediction method combining frequency domain attention mechanism and trend decomposition linear network （FDA-TDLN）was proposed. First，an FDA prediction module anda TDLN prediction module were innovatively designed to model the trend information，complex periodic characteristics and iregular fluctuations in lost circulation data.Second，through trend decompositionand frequency domain analysis of time series data，key features were effectively extracted by the model.Third，bycombining theoutputs ofthetwo modules in the formof weighted fusion，thelong-termtrend modeling capabilityoflinear network was captured while enhancing the nonlinearfeature expression abilitythrough FDA.Finally，to verify the effectivenessof the method，experimental studies were conducted on multiple lost circulation datasets，and comparisons were made with existing mainstream prediction methods.The experimental results show that the method has excellnt performance in prediction accuracy，and its generalization and superiorityare validated through testing on public datasets such as ETT，Weather and Electricity.The proposed lost circulation prediction method integrating frequency domain atentionand trend decomposition provides new research ideas and technical support for solving lost circulation problems in oil production.

Keywords：lost circulation prediction； inteligent prediction；FDA； linear network；deep learning；frequency domain analysis

0 引言

井漏是指钻井液因地层渗透率过高或孔隙过大而流入地层的现象，是钻井中的常见问题。（剩余14681字）