@inproceedings{scholars11918,
         journal = {SPE/AAPG/SEG Asia Pacific Unconventional Resources Technology Conference 2019, APUR 2019},
       publisher = {Unconventional Resources Technology Conference (URTEC)},
           title = {Production forecasting for shale gas well in transient flow using machine learning and decline curve analysis},
            year = {2019},
             doi = {10.15530/ap-urtec-2019-198198},
            note = {cited By 13; Conference of SPE/AAPG/SEG Asia Pacific Unconventional Resources Technology Conference 2019, APUR 2019 ; Conference Date: 18 November 2019 Through 19 November 2019; Conference Code:157342},
            isbn = {9781613996737},
          author = {Han, D. and Kwon, S. and Son, H. and Lee, J.},
        abstract = {Decline curve analysis (DCA) is widely used to predict the future productivity for a transient flow (TF) regime in a shale gas well. However; when a production profile and productivity are estimated by applying a single equation, it is not easy to determine an appropriate DCA method and produce accurate estimations. Among various parameters, which have been proposed to improve the problem of selecting an appropriate DCA method, the reservoir permeability is not effective since this data is not consistently available datay. In addition, the decline rate index technique has also limitations because a decline rate becomes irregular due to the suspended production during the maintenance of a well development site. As for accurate production forecast, the existing DCA methods accurately represent a future production profile for the transient flow regime, but these methods tend to over-or under-estimate performance during the boundary-dominated flow (BDF). This study developed a data-based artificial neural network (ANN) model for predicting future production rates during the transient flow regime when using a DCA method. Input data for the ANN model are hydraulic fracture factor, well completion factor, reservoir properties and the production management data. Output data are the future production rates. A total of 150 well cases comprised of 8560 individual data points were used for training. To improve the prediction performance of the model, a machine learning technique was utilized to design optimal input and output data and the effectiveness of the model was verified by conducting a clustering analysis that identified similarity among datasets. We performed DCA on the historical production rates, and the future production rates and cumulatives calculated from the proposed ANN model were predicted to be at least 1 and up to 8.1 more accurate than when using only the existing production rates. {\^A}{\copyright} 2019, Unconventional Resources Technology Conference (URTeC).},
        keywords = {Forecasting; Hydraulic fracturing; Machine learning; Natural gas wells; Neural networks; Petroleum reservoir engineering; Productivity; Reservoir management; Resource valuation; Shale gas; Transition flow; Well completion, Artificial neural network models; Boundary dominated flow; Decline curve analysis; Machine learning techniques; Prediction performance; Production forecasting; Production management; Reservoir permeability, Natural gas well production},
             url = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-85085667012&doi=10.15530\%2fap-urtec-2019-198198&partnerID=40&md5=f8fd701fd7bf60136f9cbe48b1f650a6}
}