A Study Of Low Permeability Oilfield Reservoir Evaluation Method Based On Machine Learning

With the acceleration of offshore oil development in China,the evaluation of low permeability oil and gas reservoirs becomes a main research direction in the field of oil and gas exploration.Low permeability reservoirs have complex pore structures and intricate geological structures;and its distribution of oil and gas resources is uneven.Therefore,its reservoir evaluation requires comprehensive reservoir parameters,and fully considers the spatial correlation between pre-sampling points and post-sampling points to improve the accuracy of parameter prediction.Besides,it takes large funds and lots of manpower to develop oil and gas in low permeability reservoirs,which means if the reservoir parameters used to classify and evaluate the reservoirs are not accurate enough,we may unfortunately witness resources overconsumption,fund waste and environment losses.In response to these issues,this study takes the M oilfield as the research object,and adopts machine learning technology and the neural network model derived from it.Based on solid investigations of various reservoir parameter prediction and classification evaluation methods,the study uses dynamic reservoir quality index as a key measure.This paper designs a set of machine learning-based evaluation methods for low permeability oilfield reservoirs based on three aspects(logging parameter optimization,dynamic reservoir quality index prediction and reservoir classification evaluation),and establishes a supporting integrated management system based on the above methods.The main research works are as follows:1.In order to select the most ideal combination of logging parameters and dynamic reservoir quality index,this study adopts the extreme gradient boosting algorithm,and evaluates the feature sensitivity by calculating the split contribution of the feature in the decision tree,sample weight and feature split threshold.The sensitivity parameters of the dynamic reservoir quality index are obtained based on sensitivity ranking to select the optimal logging parameters.2.After fully considering the correlations between different characteristic parameters and the variation of logging parameters with depth,this study proposes a neural network model combining one-dimensional convolution(1DCNN)and bidirectional gated recurrent unit(BiGRU)based on the optimal logging parameter combination.This model is expected to improve the prediction accuracy of the dynamic reservoir quality index,and the results are analyzed and compared with the BP and one-dimensional convolutional neural network models to verify the performance advantages of the composite model.3.In order to solve the problem of large physical property differences between different layers of low-permeability reservoirs,this study uses a multi-resolution automatic clustering algorithm based on graph theory,combining natural gamma ray and dynamic reservoir quality index,to propose a unified low-permeability reservoir classification evaluation method.Comprehensive analysis can be done based on the differences between geological characteristics and logging response characteristics to find the abundant oilfield layers of the reservoir in the study area.4.In order to put research into practice,this study extends the application of the proposed algorithm,designs and develops an integrated parameter prediction and reservoir evaluation management system;and finally,the practical value and significance of the theory are validated through the application of the system.The application and experimental results show that the machine learning-based low permeability oilfield reservoir evaluation method combined with the dynamic reservoir quality index comprehensively investigates the correlation between depth and logging parameters.Besides,the study formulates a common standard of classification evaluation method for different reservoir layers which effectively resolves the complexity of low permeability reservoirs,and improves prediction accuracy.The evaluation method provides new theoretical and technical reserves for finding abundant reservoirs,which is of great value in practice.