Font Size: a A A

EUR Prediction Of Shallow Shale Gas Based On Machine Learning Method

Posted on:2023-07-11Degree:MasterType:Thesis
Country:ChinaCandidate:Y H MaFull Text:PDF
GTID:2531307163989099Subject:Oil and gas field development project
Abstract/Summary:PDF Full Text Request
The shallow shale gas exploration breakthrough in the Taiyang anticline area of Zhaotong was achieved from 2017 to 2018.The area and resources of shallow shale gas favorable areas less than 2000 m accounted for nearly half of the total demonstration area.The success shows that the shallow shale gas in the southern marine residual basin has huge exploration prospect and development potential.Taiyang Shallow shale gas has complex structure,shallow burial depth,insufficient formation energy,developed faults and natural fractures,and the main layers of the Longmaxi Wufeng Formation are characterized by small continuous thickness of the Long 11 and Long 12-1 sublayers.The production dynamics of gas wells are complex,the production fluctuates greatly,and the decline characteristic is not obvious.The reservoir and production characteristics of shallow shale gas in the Taiyang block have brought great difficulties to traditional methods for predicting and estimating recoverable reserves EUR.The model workload is large,and the reservoir numerical simulation has high requirements on the accuracy of the geological model,the modeling period is long,and there is a hysteresis.In order to solve the problems existing in traditional oil and gas reservoir engineering methods in predicting EUR,this thesis compares the applicability of three machine learning methods in predicting EUR of shallow shale gas,preprocesses production data,and finds that BP neural network and LSTM has a fairly good effect in predicting the decrement curve and the cumulative yield curve and calculating EUR,and can predict EUR quickly and accurately;as a neighborhood-based method,random forest cannot predict the decrement and cumulative yield,and is not applicable.At the same time,this thesis proposes a data-physical hybrid driving model to predict the EUR of a single well.On the basis of further improving the accuracy of EUR prediction for shallow shale gas,the complexity of the model is guaranteed to be low,and the EUR can be quickly performed.Prediction,clarify the development potential of shale gas wells,and guide the development of shallow shale gas.
Keywords/Search Tags:EUR, Shallow, Shale Gas, Machine Learning, Hybrid driven
PDF Full Text Request
Related items