Reservoir Identification And Physical Property Parameter Prediction Based On Machine Learning

Reservoir is the place of oil and gas storage,and it is also the direct target layer of oil and gas field development.It is an important research work in oil and gas field exploration and development to quickly and accurately identify the reservoir and predict the physical parameters of the reservoir.The existing reservoir logging identification methods fail to consider the time series relationship of logging data,and the identification effect is poor,and there are certain limitations.To this end,deep learning is applied to reservoir identification.Deep learning has powerful nonlinear mapping capabilities and the ability to express data features,and can be used to deal with arbitrary linear and nonlinear relationships.Based on the logging data of the production wells in a well area of Dagang Oilfield,this thesis selects four sets of logging data,that are more sensitive to the reservoir type through investigation,reasonable neural network processing layers and reasonable neural network parameters and establishes the LSTM neural network,GRU neural network,BILSTM neural network,BIGRU neural network reservoir identification model,predict the reservoir type of other unknown wells in this well area.Finally,the results of the four predictions are compared and analyzed,and the best deep neural network is obtained and some improvements are made through experiments.For the prediction of the physical parameters of the reservoir,this thesis uses three commonly used machine learning regression algorithms: decision tree,support vector machine and Gaussian process regression to regress the reservoir porosity,permeability and oil saturation.In order to find the best regression model,the author replacs different parameters and kernel functions.The regression results show that the BIGRU deep neural network shows excellent recognition performance in reservoir recognition,and the recognition accuracy reaches 93.22 %,which is 3 %～4 % higher than that of LSTM,GRU and BILSTM.The improved BIGRU recognition accuracy is 95.02 %.Exponential kernel Gaussian process regression is the best porosity regression model,quadratic rational kernel Gaussian process regression shows excellent performance in the regression of permeability and oil saturation.This article provides more information for future reservoir identification and physical parameter prediction.