Training Image Optimization Method Based on Convolutional Neural Network and Its Application in Discrete Fracture Network Model Selection

Abstract Training image (TI) is important for multipoint statistics simulation method (MPS), since it captures the spatial geological pattern of target reservoir to be modeled. Generally, one optimal TI selected before applying MPS by evaluating similarities between many TIs and well interpretations reservoir. In this paper, we propose a new training optimization approach based on convolutional neural network (CNN). First, candidate were randomly sampled several times obtain sample dataset. Then, CNN was used conduct transfer learning all samples, finally, conditioning data through trained model. By taking advantage strong ability in feature recognition, proposed can automatically identify differences features samples image. Hence, effectively resolves difficulty matching discrete datapoints grid structures. We demonstrated applicability our model via 2D 3D selection examples. The methods appropriate TI, then pretreatment techniques improving accuracy continuous achieved. Moreover, successfully applied fracture Finally, sensitivity analysis carried out show that sufficient volume reduce uncertainty results. comparing with improved MDevD method, advantages are verified terms efficiency reliability.