Oil reservoir automatic history fitting method based on generative adversarial network

Abstract

The invention provides an oil reservoir automatic history fitting method based on a generative adversarial network, and the method comprises the steps: carrying out the collection of data, and forminga data set through permeability field data and production data; preprocessing the data set, and dividing the data set into a training set, a verification set and a test set; training a generative adversarial network model by using the training set data, and selecting an optimal model for storage by using the verification set data; reading the stored optimal model, inputting test set data, and generating a corresponding permeability field; performing numerical simulation on the permeability field data generated in the previous step to obtain production data b, comparing the production data b with production data B corresponding to a real permeability field in the test set, and performing result verification; and applying the model which is verified to be qualified through a result to actual oil well geological condition prediction. According to the method, an agent model or expert experience is not needed, the mapping relation between static parameters (permeability fields) and dynamicproduction data is found by training the generative adversarial network, and the geological condition is obtained through the oil well production data.

Description

technical field [0001] The invention belongs to the technical field of automatic history fitting methods for oil reservoirs, in particular to an automatic history fitting method for oil reservoirs based on generative confrontation networks. Background technique [0002] With the continuous improvement of computer data processing capabilities and the gradual maturity of deep learning algorithms, deep learning algorithms have begun to be applied to practical applications. Reservoir automatic history matching inverts the subsurface geological conditions by using dynamic production data such as fluid production of oil wells and water wells or relative permeability lines. An effective history fitting method can be regarded as a continuous fitting process to the actual production data. When the production curve corresponding to the corrected prior model is consistent with the actual production curve, the inversion result is obtained. [0003] Most of the traditional history match...

AI Technical Summary

Problems solved by technology

[0004] The present invention provides an automatic history fitting method for reservoirs based on generative confrontation network, which overcomes the problems that traditional history fitting algorithms are mostly based on Gaussian distribution, it is difficult to converge complex reservoir models, and it is difficult to obtain actual underground models

Images