Lu Qiao, Shengyu Yang, Qinhong Hu, Huijun Wang, Taohua He
{"title":"无监督对比学习:基于常规测井的页岩孔隙度预测","authors":"Lu Qiao, Shengyu Yang, Qinhong Hu, Huijun Wang, Taohua He","doi":"10.1063/5.0206449","DOIUrl":null,"url":null,"abstract":"Porosity is a pivotal factor affecting the capacity for storage and extraction in shale reservoirs. The paucity of labeled data in conventional well logs interpretation and supervised learning models leads to inadequate generalization and diminished prediction accuracy, thus limiting their effectiveness in precise porosity evaluation. This study introduces a contrastive learning – convolutional neural network (CL-CNN) framework that utilizes CL for pretraining on a vast array of unlabeled data, followed by fine-tuning using a traditional CNN on a curated set of labeled data. Applied to the Subei Basin in Eastern China, the framework was tested on 130 labeled data and 2576 unlabeled data points from well H1. The results indicate that the CL-CNN framework outperforms traditional CNN-based supervised learning and other machine learning models in terms of prediction accuracy for the dataset under consideration. Furthermore, it demonstrates the potential for extensive porosity assessment across different logged depths. Due to its efficacy and ease of use, the proposed framework is versatile enough for application in reservoir evaluation, engineering development, and related fields. The innovative contribution of this research is encapsulated in its unique methodology and procedural steps for the accurate prediction of shale reservoir porosity, thus significantly enriching the existing body of knowledge in this domain.","PeriodicalId":509470,"journal":{"name":"Physics of Fluids","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2024-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Unsupervised contrastive learning: Shale porosity prediction based on conventional well logging\",\"authors\":\"Lu Qiao, Shengyu Yang, Qinhong Hu, Huijun Wang, Taohua He\",\"doi\":\"10.1063/5.0206449\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Porosity is a pivotal factor affecting the capacity for storage and extraction in shale reservoirs. The paucity of labeled data in conventional well logs interpretation and supervised learning models leads to inadequate generalization and diminished prediction accuracy, thus limiting their effectiveness in precise porosity evaluation. This study introduces a contrastive learning – convolutional neural network (CL-CNN) framework that utilizes CL for pretraining on a vast array of unlabeled data, followed by fine-tuning using a traditional CNN on a curated set of labeled data. Applied to the Subei Basin in Eastern China, the framework was tested on 130 labeled data and 2576 unlabeled data points from well H1. The results indicate that the CL-CNN framework outperforms traditional CNN-based supervised learning and other machine learning models in terms of prediction accuracy for the dataset under consideration. Furthermore, it demonstrates the potential for extensive porosity assessment across different logged depths. Due to its efficacy and ease of use, the proposed framework is versatile enough for application in reservoir evaluation, engineering development, and related fields. The innovative contribution of this research is encapsulated in its unique methodology and procedural steps for the accurate prediction of shale reservoir porosity, thus significantly enriching the existing body of knowledge in this domain.\",\"PeriodicalId\":509470,\"journal\":{\"name\":\"Physics of Fluids\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-05-01\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Physics of Fluids\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1063/5.0206449\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Physics of Fluids","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1063/5.0206449","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
Unsupervised contrastive learning: Shale porosity prediction based on conventional well logging
Porosity is a pivotal factor affecting the capacity for storage and extraction in shale reservoirs. The paucity of labeled data in conventional well logs interpretation and supervised learning models leads to inadequate generalization and diminished prediction accuracy, thus limiting their effectiveness in precise porosity evaluation. This study introduces a contrastive learning – convolutional neural network (CL-CNN) framework that utilizes CL for pretraining on a vast array of unlabeled data, followed by fine-tuning using a traditional CNN on a curated set of labeled data. Applied to the Subei Basin in Eastern China, the framework was tested on 130 labeled data and 2576 unlabeled data points from well H1. The results indicate that the CL-CNN framework outperforms traditional CNN-based supervised learning and other machine learning models in terms of prediction accuracy for the dataset under consideration. Furthermore, it demonstrates the potential for extensive porosity assessment across different logged depths. Due to its efficacy and ease of use, the proposed framework is versatile enough for application in reservoir evaluation, engineering development, and related fields. The innovative contribution of this research is encapsulated in its unique methodology and procedural steps for the accurate prediction of shale reservoir porosity, thus significantly enriching the existing body of knowledge in this domain.