Research On Logging Curve Super-resolution Method For Fine Description Of Shale Reservoirs

With the exploration and development of larger oil and gas fields in China and around the world in the mid to late stages,the current exploration and development goals are changing from traditional oil and gas resources to unconventional resources.However,due to the impact of the tense international relations in today’s society and the wanton spread of the global new crown virus on the global oil and gas supply and demand,the price of crude oil is at the highest value in the past period of time.In this global environment,further increasing domestic energy production capacity and energy production quality is inevitable to ensure the lasting and longterm stable development of the national economy and the steady improvement of national defense strength.Fine reservoir description plays a very important role in improving the quality of oil and gas production,and research work by improving the resolution of logging data is an important link in fine reservoir description.Therefore,this paper draws on the concepts and practices related to super-resolution in the field of image and audio.Modern signal processing theory,machine learning methods,and deep learning theory methods are used in combination to train and learn from logging data.The nonlinear mapping relationship between highresolution data and low-resolution data was established through the application of cross-well data and the effective mining of longitudinal logging information,the research work on the super-resolution method of logging curves for the fine description of reservoirs has been carried out.The existing high-resolution processing methods for logging curves are no longer suitable for resolution enhancement in the context of big data,as they are unable to achieve effective resolution enhancement,and based on this,we propose a study on the super-resolution method of local linear embedding of logging curves.The method implementation combines the concept of super-resolution among image professional fields,aiming at mining the potential joint application relationships of inter-well related data.Specifically,by assuming that the highresolution and low-resolution curvature curves are in the same popular space and they have similar structures,the non-linear mapping relationship between high-and low-resolution curves can be found by training and learning from different inter-well data,and finally the resolution enhancement task of low-resolution curves at 2x and 4x scales can be completed.In combination with several experimental comparisons,visually as well as in the evaluation metrics,the super-resolution processing task of the logging curve achieved by this method substantially improves the recovery accuracy compared to other high-resolution processing methods and can achieve a remarkable result.To address the shortcomings of the above proposed method in the poor performance of some curves(e.g.SP curve)and the inability to train and then super-resolve all curve data between the same well,this chapter proposes a study of the LSTM-RF method for superresolution of logging curves based on this.This method can deeply mine the potential relationship between different curves of the same well by using the excellent ability of LSTM to mine and process sequence data.Specifically,upsampling and downsampling operations are performed on conventional logging curves to generate data of different scales,and then multiview conventional interpolation is performed to obtain the initial high-resolution data sequence,which is then input to the LSTM network for correlation,and the nonlinear mapping relationship is obtained.Through the experimental process of actual logging data,it can be seen that this method has better super-resolution effect,lower root mean square error value,make the most of LSTM’s excellence,and has particularly strong stability.