Current Location:home > Browse
Your conditions: Latest Submit(1)

1. chinaXiv:202104.00062 [pdf]

Training Image Optimization Method Based on Convolutional Neural Network

Siyu, YU; Shaohua, LI
Subjects: Geosciences >> Other Disciplines of Geosciences

As a prior geological model, which can effectively describe the internal structure of the reservoir, geometry and distribution of sedimentary facies, training image(TI) is the key input data of multipoint statistics(MPS). Before implementing MPS modeling, appropriate TIs must be provided to match the spatial correlation of the work area condition data. The essence of choosing the training image that best matches the conditional data is to quantitatively evaluate the similarity of spatial features between the discrete points and the regular grid. This paper presents a new training image optimization method based on Convolutional Neural Network(CNN). Firstly, candidate TIs were randomly sampled several times to obtain the sample point set. Then, CNN was used to conduct transfer learning for all samples. Finally, the optimal TI of the conditional data is selected through the trained CNN model. Based on the strong learning ability of CNN in image feature recognition, the proposed method can automatically identify the spatial feature differences between the conditional data and the sample points of the training image, and effectively solves the difficulty of spatial matching between discrete points and grid structures. Through four different TI optimization cases, it is proved that the new method can effectively select suitable TI. In the discussion, we put forward a trick to improve the accuracy of continuous TI recognition by discretizing continuous variables. Finally, sensitivity analysis was carried out, and it was found that sufficient conditional data was the key factor to select suitable TI.

submitted time 2021-04-14 Hits76Downloads29 Comment 0

  [1 Pages/ 1 Totals]