论文详情
识别裂缝是油气储量评价和产能预测的关键。目前识别裂缝主要采用基于人机交互的方法,该方法耗功耗时且易受主观因素影响,因而对裂缝的识别不够精确。为此,提出利用条件生成对抗网络(CGAN)识别图像中的裂缝。CGAN通过训练给定的图像和对应标签图像,提取训练图像和标签中的特征,以此特征识别图像中的信息。利用CGAN识别模拟图像中的正弦形态裂缝,识别裂缝准确率达93.4%。CGAN对地层微电阻率扫描成像(FMI)图像中的水平缝和低角度缝识别准确率为90%。研究结果表明,和蚁群算法相比,CGAN是一种效果好、速度快及抗干扰能力强的计算机自动识别裂缝方法。
Fracture identification is key to both the suitable evaluation of oil and gas reserves and an accurate prediction of productivity.At present,fracture identification is primarily based on methods involving human-computer interaction,which are time-consuming and vulnerable to subjective factors; this leads to the inaccurate identification of fractures.In this paper,the conditional generative adversarial nets (CGAN) method is proposed to identify fractures in images.By training the method on given images and corresponding label images,relevant features can be extracted to achieve the desired identification.This study used the CGAN method to identify sinusoidal fractures in simulated images,achieving an accuracy of 93.4%.The method was also applied on horizontal and low-angle fractures on a Formation Micro-Scanner Image (FMI),achieving an accuracy of 90%.Compared with the results of an ant colony algorithm,the CGAN was found to be more effective,efficient,and exhibited better anti-jamming capability.
吉林省科技发展优秀青年人才项目(20190103150JH)资助。