Research On Prediction Method Of Beach Bar Sand Reservoir Based On Machine Learning

With the development of machine learning technology,this emerging technology has been applied to the production practice of petroleum exploration and development.The beach bar sand reservoir of Y Oilfield has the characteristics of high concealment,low data resolution,and large lateral variation of the reservoir sand body compared with the channel sand reservoir.Based on the current basic data and traditional technology,the reservoir prediction accuracy cannot be met Needs,so this paper attempts to apply machine learning methods to the beach bar sand reservoir,comprehensive analysis and comparison of random forest model,BP neural network model,decision tree model to predict the distribution of reservoir sand body.First,make full use of the basic data containing the geological information of the work area,pre-process the data in the research block,including the normalization processing of the logging curve,seismic data volume interpolation,time-depth calibration,etc.,and then divide the frequency of the selected seismic attributes.Then use the five curves GR,AC,R4,R25,and COND to select seismic attributes sensitive to reservoir parameters.Then,using BP neural network,decision tree and random forest modeling,the random forest model with R-Squared value of 0.66 is comprehensively selected based on the indicators of R-Squared and RMSE to calculate the attribute mapping body of these five logging curves.Calculate the importance of the five logging curves to lithology again,and use this importance as the weight to fuse the five logging attribute bodies into a data body to obtain the attribute fusion body.Compared with the GR data volume profile,the seismic data profile can observe a clearer reservoir sand body boundary,a more obvious sand body position,and a better prediction effect.The method used in this article is less human intervention,faster calculation speed,more objective and true.