Cheng Jian-yong, Yuan San-yi, Sun Ao-xue, Luo Chun-mei,*, Liu Hao-jie, and Wang Shang-xu
1. State Key Laboratory of Petroleum Resources and Engineering, China University of Petroleum (Beijing), Beijing, China.102249
2. PetroChina Changqing Oilfi eld Company No.8 Oil Production Company, Xian 718600, China
3. Shengli Geophysical Research Institute of Sinopec, Dongying 257000, China.
Abstract：
Traditional deep learning methods pursue complex and single network architectures without considering the petrophysical relationship between different elastic parameters. The mathematical and statistical significance of the inversion results may lead to model overfitting, especially when there are a limited number of well logs in a working area. Multitask learning provides an effective approach to addressing this issue. Simultaneously, learning multiple related tasks can improve a model’s generalization ability to a certain extent, thereby enhancing the performance of related tasks with an equal amount of labeled data. In this study, we propose an end-to-end multitask deep learning model that integrates a fully convolutional network and bidirectional gated recurrent unit for intelligent prestack inversion of “seismic data to elastic parameters.” The use of a Bayesian homoscedastic uncertainty-based loss function enables adaptive learning of the weight coeffi cients for different elastic parameter inversion tasks, thereby reducing uncertainty during the inversion process. The proposed method combines the local feature perception of convolutional neural networks with the long-term memory of bidirectional gated recurrent networks. It maintains the rock physics constraint relationships among different elastic parameters during the inversion process, demonstrating a high level of prediction accuracy. Numerical simulations and processing results of real seismic data validate the effectiveness and practicality of the proposed method.
2025, Vol. 22 
