CN-122021296-A - Reservoir modeling task processing method based on hybrid domain parallel perception generation countermeasure network

CN122021296ACN 122021296 ACN122021296 ACN 122021296ACN-122021296-A

Abstract

The application discloses a reservoir modeling task processing method based on hybrid domain parallel perception generation countermeasure network, which relates to the technical field of geostatistics, and aims to solve the problems of large training data demand and insufficient feature pattern mining capability by utilizing a cross-domain migration driven lightweight reservoir configuration sample enhancement model to carry out parameter migration and expansion on original sample data on the basis of training parameters of existing samples, introducing a parallel attention mechanism supporting multi-source features and a multi-scale feature deep fusion mechanism by utilizing a multi-dimensional hybrid domain joint characterization module to construct a countermeasure network based on hybrid domain parallel perception fusion, inputting sample data into the generated countermeasure network, outputting target data, carrying out consistency comparison on the target data and an expanded sample data set, and processing reservoir configuration modeling tasks if the comparison is passed and the model evaluation of the generated countermeasure network is passed, thereby improving the multi-scale characterization capability of the model and improving the model applicability and accuracy in complex actual scenes.

Inventors

LI WEI
YUE DALI
LIU LEI
WANG WURONG
LI ZHIBO
XU ZHENHUA
ZHONG QIAN
LI QING

Assignees

中国石油大学(北京)

Dates

Publication Date: 20260512
Application Date: 20260127

Claims (10)

1. A reservoir modeling task processing method based on hybrid domain parallel perception generation countermeasure network is characterized by comprising the following steps: Based on a convolution block attention module, constructing a cross-domain migration driven lightweight reservoir configuration sample enhancement model, performing parameter migration and expansion on original sample data meeting second data volume on the basis of training parameters of existing samples meeting first data volume to obtain an expanded sample data set, and determining sample data based on the expanded sample data set; The method comprises the steps of utilizing a multi-dimensional mixed domain joint characterization module, and introducing a parallel attention mechanism supporting multi-source features and a multi-scale feature deep fusion mechanism to construct a parallel perception fusion based on a mixed domain to generate an countermeasure network, wherein the multi-dimensional mixed domain joint characterization module is a module for processing sample data in terms of a space domain and a frequency domain; Inputting the sample data into a generation countermeasure network to output target data, and comparing the target data with the extended sample data set in a consistency manner; if the consistency comparison is passed, carrying out model evaluation on the generated countermeasure network, and if the model evaluation is passed, taking the generated countermeasure network as a target to generate the countermeasure network; And acquiring a reservoir configuration modeling task, and processing the reservoir configuration modeling task by utilizing the target generation countermeasure network.
2. The hybrid domain parallel perception fusion-based intelligent hybrid domain parallel perception generation countermeasure network-based reservoir modeling task processing method according to claim 1, wherein the constructing a cross-domain migration-driven lightweight reservoir configuration sample enhancement model based on a convolution block attention module comprises: and introducing an attention mechanism at a jump connection part among all the convolution block attention modules to construct a lightweight reservoir configuration sample enhancement model of the cross-domain migration drive, wherein the convolution block attention modules comprise a channel attention module and a space attention module, and the reservoir configuration sample enhancement model is a lightweight network of the cross-domain migration drive.
3. The method for processing reservoir modeling tasks based on hybrid domain parallel perception generation countermeasure network according to claim 1, wherein the performing parameter migration and expansion on original sample data satisfying a second data volume based on training parameters of existing samples satisfying a first data volume to obtain an expanded sample data set, determining sample data based on the expanded sample data set, comprises: Inputting original sample data meeting the second data volume into a cross-domain migration driven lightweight reservoir configuration sample enhancement model so that the reservoir configuration sample enhancement model performs parameter migration and expansion on the original sample data on the basis of training parameters of the existing samples meeting the first data volume to obtain a model and an expansion sample data set which accord with a deposition mode; Sample data is determined based on the model conforming to the deposition pattern and the extended sample data set.
4. The method of processing reservoir modeling tasks based on hybrid domain parallel perception generation countermeasure network of claim 1, wherein the inputting the sample data into the generation countermeasure network to output target data includes: And inputting the sample data into a generating countermeasure network so as to process the sample data in terms of a space domain and a frequency domain by utilizing a multidimensional mixed domain joint characterization module in the generating countermeasure network, introducing a parallel attention mechanism supporting multi-source features and a multi-scale feature deep fusion mechanism, and performing data fitting and feature mapping on the sample data to output target data.
5. The method for processing reservoir modeling tasks based on hybrid domain parallel awareness generated countermeasure network of claim 4, wherein the processing of the sample data in terms of spatial domain and frequency domain with the multi-dimensional hybrid domain joint characterization module in the generated countermeasure network comprises: Processing the sample data in terms of local detail dimension and global structure dimension by utilizing the multi-dimensional mixed domain joint characterization module in the generation countermeasure network; Utilizing the multi-dimensional mixed domain joint characterization module in the generation countermeasure network, extracting spatial characteristics based on maximum pooling and average pooling operation, and performing spatial domain processing on the sample data; And utilizing the multidimensional mixed domain joint characterization module in the generation countermeasure network, and carrying out frequency domain processing on the sample data based on a haar wavelet transformation method.
6. The hybrid domain parallel awareness generation countermeasure network based reservoir modeling task processing method of claim 1, wherein the consistency comparison of the target data and the extended sample data set includes: Screening standard data from the extended sample data set; comparing the error between the standard data and the target data by using a discriminator; and if the error meets a preset comparison passing condition, judging that the consistency comparison passes.
7. Reservoir modeling task processing method based on hybrid domain parallel perception generated countermeasure network according to any of claims 1 to 6, characterized in that the model evaluation of the generated countermeasure network includes: determining model evaluation indexes, wherein the model evaluation indexes comprise a phase proportion, a multi-scale transformation scaling function in the horizontal axis direction and a variation function in the vertical axis direction; And carrying out model evaluation on the generated countermeasure network based on the model evaluation index.
8. A reservoir modeling task processing device for generating an countermeasure network based on hybrid domain parallel awareness, comprising: The reservoir configuration sample enhancement model construction module is used for constructing a cross-domain migration driven lightweight reservoir configuration sample enhancement model based on the convolution block attention module, performing parameter migration and expansion on original sample data meeting second data volume on the basis of training parameters of existing samples meeting first data volume so as to obtain an expanded sample data set, and determining sample data based on the expanded sample data set; The system comprises a generation countermeasure network construction module, a multi-dimensional mixed domain joint characterization module, a multi-dimensional mixed domain analysis module and a comparison module, wherein the generation countermeasure network construction module is used for utilizing the multi-dimensional mixed domain joint characterization module and introducing a parallel attention mechanism supporting multi-source characteristics and a multi-scale characteristic deep fusion mechanism to construct a countermeasure network based on mixed domain parallel perception fusion; The consistency comparison module is used for inputting the sample data into a generated countermeasure network so as to output target data, and carrying out consistency comparison on the target data and the extended sample data set; The model evaluation module is used for carrying out model evaluation on the generated countermeasure network if the consistency comparison is passed, and taking the generated countermeasure network as a target to generate the countermeasure network if the model evaluation is passed; And the task processing module is used for acquiring a reservoir configuration modeling task and processing the reservoir configuration modeling task by utilizing the target generation countermeasure network.
9. An electronic device, comprising: A memory for storing a computer program; a processor for executing the computer program to implement the reservoir modeling task processing method based on hybrid domain parallel awareness generation of a countermeasure network as claimed in any of claims 1 to 7.
10. A computer readable storage medium for storing a computer program, wherein the computer program when executed by a processor implements a reservoir modeling task processing method based on hybrid domain parallel awareness generation countermeasure network as claimed in any of claims 1 to 7.

Description

Reservoir modeling task processing method based on hybrid domain parallel perception generation countermeasure network Technical Field The invention relates to the technical field of geostatistics, in particular to a reservoir modeling task processing method based on hybrid domain parallel perception generation countermeasure network. Background The existing reservoir architecture modeling technology relies on massive high-quality samples for model training, however, in actual geological engineering application, obtaining sufficient and representative real underground core or outcrop samples mainly faces the technical bottlenecks of high cost and great difficulty. Although a configuration modeling scheme based on a single sample is tried in the prior art, the feature extraction capability is limited, the generated result is seriously coupled with the style and the content of an original image, the diversified reconstruction of the configuration mode is difficult to realize, and noise points and artifacts are extremely easy to introduce. In addition, the underground reservoir configuration is complex in distribution, strong in heterogeneity and multi-level feature mining, the reservoir modeling precision is improved, the traditional convolutional neural network architecture is limited by a local receptive field, long-range dependency relations existing in reservoir space are difficult to capture effectively, and global feature mining of key geological modes is insufficient. Meanwhile, the existing modeling method is mostly limited to mining of spatial domain information, and abundant detail information contained in a frequency domain is ignored, so that feature characterization is insufficient. From the above, how to fully capture the multi-scale configuration mode features comprehensively, solve the problems of large demand of training data and insufficient feature mode mining capability, promote the multi-scale characterization capability of the model on complex structures and modes, ensure model diversity, improve the applicability and accuracy of the model in complex actual scenes, and solve the problem in the field of reservoir configuration modeling task processing. Disclosure of Invention In view of the above, the invention aims to provide a reservoir modeling task processing method based on hybrid domain parallel perception generation countermeasure network, which can fully capture multi-scale configuration mode characteristics, solve the problems of large training data demand and insufficient characteristic mode mining capability, promote multi-scale characterization capability of a model on complex structures and modes, ensure model diversity, improve model applicability and accuracy in complex actual scenes, and improve efficiency of reservoir configuration modeling task processing. The specific scheme is as follows: in a first aspect, the application discloses a reservoir modeling task processing method based on hybrid domain parallel perception generation countermeasure network, comprising the following steps: Based on a convolution block attention module, constructing a cross-domain migration driven lightweight reservoir configuration sample enhancement model, performing parameter migration and expansion on original sample data meeting second data volume on the basis of training parameters of existing samples meeting first data volume to obtain an expanded sample data set, and determining sample data based on the expanded sample data set; The method comprises the steps of utilizing a multi-dimensional mixed domain joint characterization module, and introducing a parallel attention mechanism supporting multi-source features and a multi-scale feature deep fusion mechanism to construct a parallel perception fusion based on a mixed domain to generate an countermeasure network, wherein the multi-dimensional mixed domain joint characterization module is a module for processing sample data in terms of a space domain and a frequency domain; Inputting the sample data into a generation countermeasure network to output target data, and comparing the target data with the extended sample data set in a consistency manner; if the consistency comparison is passed, carrying out model evaluation on the generated countermeasure network, and if the model evaluation is passed, taking the generated countermeasure network as a target to generate the countermeasure network; And acquiring a reservoir configuration modeling task, and processing the reservoir configuration modeling task by utilizing the target generation countermeasure network. Optionally, the constructing a cross-domain migration driven lightweight reservoir configuration sample enhancement model based on the convolution block attention module includes: and introducing an attention mechanism at a jump connection part among all the convolution block attention modules to construct a lightweight reservoir configuration sample enhancement model of the cross