H. Darabi, Xiang Zhai, A. Kianinejad, Zheren Ma, D. Castineira, R. Toronyi
{"title":"增强型人工智能框架用于非常规油藏油井动态预测和机会识别","authors":"H. Darabi, Xiang Zhai, A. Kianinejad, Zheren Ma, D. Castineira, R. Toronyi","doi":"10.2523/iptc-20099-ms","DOIUrl":null,"url":null,"abstract":"\n Many important business decisions and planning in unconventional reservoirs rely on a reliable forecast on well performance. Common practices like statistical type curves, analytical methods, and numerical simulation are not well suited to incorporate all the complexities involving rock/fluid properties, geological parameters, artificial lift systems, well and completion designs, etc. In this work, we introduce a novel \"Augmented AI\" (Artificial Intelligence) workflow for reliable forecasting of unconventional well performance and show its impact on decision making. Augmented AI represents smart integration of artificial intelligence and domain knowledge. In the application of well performance forecast, a smart DCA algorithm automatically estimates the short- and long-term performance of the historical wells; a spectrum of well attributes are aggregated/transformed with the consideration of uncertainty and robustness for training and prediction. Boosting and bootstrap tree-based models are ensembled to maximize the model generalization capability. In contrast to the commonly seen black-box modeling practices, the factor-specific impacts are deconvoluted, allowing for validation of the underlying physics. Furthermore, this gives guidelines for future well planning and completion designs. A case study is presented, where the workflow is implemented. Multi-disciplinary data (logs, completions, maps, fluid properties, etc.) from thousands of wells were integrated. During the feature engineering step, raw data was converted to a set of meaningful parameters leveraging the domain knowledge. As an example, some of the features were combined, some were transformed, and others were normalized. Then a machine learning model was created using an ensemble approach. The models showed a good model accuracy on the training, testing, and validation dataset. Leveraging the predictive model, thousands of field development opportunities including new vertical wells, new horizontal wells, recompletions, and completion optimization were identified that resulted in increased production, increased reserves, and improved capital efficiency. Using the model explanation techniques, the impact of various parameters on the well performance was quantified that resulted in best practices for future drilling and completion design.","PeriodicalId":393755,"journal":{"name":"Day 1 Mon, January 13, 2020","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2020-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"2","resultStr":"{\"title\":\"Augmented AI Framework for Well Performance Prediction and Opportunity Identification in Unconventional Reservoirs\",\"authors\":\"H. Darabi, Xiang Zhai, A. Kianinejad, Zheren Ma, D. Castineira, R. Toronyi\",\"doi\":\"10.2523/iptc-20099-ms\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"\\n Many important business decisions and planning in unconventional reservoirs rely on a reliable forecast on well performance. Common practices like statistical type curves, analytical methods, and numerical simulation are not well suited to incorporate all the complexities involving rock/fluid properties, geological parameters, artificial lift systems, well and completion designs, etc. In this work, we introduce a novel \\\"Augmented AI\\\" (Artificial Intelligence) workflow for reliable forecasting of unconventional well performance and show its impact on decision making. Augmented AI represents smart integration of artificial intelligence and domain knowledge. In the application of well performance forecast, a smart DCA algorithm automatically estimates the short- and long-term performance of the historical wells; a spectrum of well attributes are aggregated/transformed with the consideration of uncertainty and robustness for training and prediction. Boosting and bootstrap tree-based models are ensembled to maximize the model generalization capability. In contrast to the commonly seen black-box modeling practices, the factor-specific impacts are deconvoluted, allowing for validation of the underlying physics. Furthermore, this gives guidelines for future well planning and completion designs. A case study is presented, where the workflow is implemented. Multi-disciplinary data (logs, completions, maps, fluid properties, etc.) from thousands of wells were integrated. During the feature engineering step, raw data was converted to a set of meaningful parameters leveraging the domain knowledge. As an example, some of the features were combined, some were transformed, and others were normalized. Then a machine learning model was created using an ensemble approach. The models showed a good model accuracy on the training, testing, and validation dataset. Leveraging the predictive model, thousands of field development opportunities including new vertical wells, new horizontal wells, recompletions, and completion optimization were identified that resulted in increased production, increased reserves, and improved capital efficiency. Using the model explanation techniques, the impact of various parameters on the well performance was quantified that resulted in best practices for future drilling and completion design.\",\"PeriodicalId\":393755,\"journal\":{\"name\":\"Day 1 Mon, January 13, 2020\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2020-01-13\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"2\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Day 1 Mon, January 13, 2020\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.2523/iptc-20099-ms\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Day 1 Mon, January 13, 2020","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.2523/iptc-20099-ms","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
Augmented AI Framework for Well Performance Prediction and Opportunity Identification in Unconventional Reservoirs
Many important business decisions and planning in unconventional reservoirs rely on a reliable forecast on well performance. Common practices like statistical type curves, analytical methods, and numerical simulation are not well suited to incorporate all the complexities involving rock/fluid properties, geological parameters, artificial lift systems, well and completion designs, etc. In this work, we introduce a novel "Augmented AI" (Artificial Intelligence) workflow for reliable forecasting of unconventional well performance and show its impact on decision making. Augmented AI represents smart integration of artificial intelligence and domain knowledge. In the application of well performance forecast, a smart DCA algorithm automatically estimates the short- and long-term performance of the historical wells; a spectrum of well attributes are aggregated/transformed with the consideration of uncertainty and robustness for training and prediction. Boosting and bootstrap tree-based models are ensembled to maximize the model generalization capability. In contrast to the commonly seen black-box modeling practices, the factor-specific impacts are deconvoluted, allowing for validation of the underlying physics. Furthermore, this gives guidelines for future well planning and completion designs. A case study is presented, where the workflow is implemented. Multi-disciplinary data (logs, completions, maps, fluid properties, etc.) from thousands of wells were integrated. During the feature engineering step, raw data was converted to a set of meaningful parameters leveraging the domain knowledge. As an example, some of the features were combined, some were transformed, and others were normalized. Then a machine learning model was created using an ensemble approach. The models showed a good model accuracy on the training, testing, and validation dataset. Leveraging the predictive model, thousands of field development opportunities including new vertical wells, new horizontal wells, recompletions, and completion optimization were identified that resulted in increased production, increased reserves, and improved capital efficiency. Using the model explanation techniques, the impact of various parameters on the well performance was quantified that resulted in best practices for future drilling and completion design.