Research On Data Analysis Method Of Permanent Downhole Gauges Based On Machine Learning

With the rapid increase of interal demand for oil and natural gas resources,oilfield production management has put forward higher requirements for reservoir monitoring.The permanent downhole gauge(PDG)collects production data every few seconds,which can continuously monitor production condition of wellbore and reservoir for a long time and provide massive data.How to transform mass production data into useful reservoir information is a difficulty for current research.Machine learning is a branch of artificial intelligence technology in the field of computer science,providing new ideas to process and analyze production data.Based on the machine learning method,which has a good prospect for field application,the PDG production data is used for training to obtain a more accurate reservoir production prediction model and reconstruct unknown production history or predict future reservoir production status.First uses the wavelet transform denoising method to reduce the noise of the PDG data and remove the abnormal points,which provides more reliable data for the subsequent work.Then,the study reconstructs flow rate and pressure history of PDG simulation production data,UK North Sea Oilfield PDG data and Norwegian Volve field data by three machine learning methods: Linear Regression,Support Vector Machine and Long Short-Term Memory,training and learning for reservoir dynamic production,quantitatively compare the prediction capabilities of reservoir production status with three machine learning methods.Finally,the flowing bottom hole pressure history based on machine learning reconstruction is used for deconvolution well test analysis,the obtained analysis result is compared with the original production data well test analysis result to comprehensively obtain the machine learning method with the best production history reconstruction effect.The research shows that the wavelet transform denoising method can effectively reduce the noise of PDG data.Linear Regression can reconstruct the flow rate and pressure history accurately and learn condition of reservoir production dynamics quickly,which has a promising research prospects.