Research On Reservoir Parameter Prediction System Based On Deep Learning

Reservoir parameters refer to important parameters that describe the physical properties of reservoirs,such as porosity and permeability,which determine the storage and development ability of oil and gas in reservoirs.Therefore,reservoir parameter prediction is of great significance for petroleum exploration and development,geological modeling and fine evaluation of oil and gas reservoirs.Traditional reservoir parameter calculation methods are to use logging data to establish empirical formulas or simplified geological models.However,these methods only consider the linear relationship between variables and ignore the characteristics of heterogeneity and anisotropy of the reservoir,and their prediction accuracy is low.How to build a prediction model that can reflect the nonlinear mapping relationship between logging data and reservoir parameters is particularly important.To solve the above problems,the deep learning method is applied to the prediction of reservoir parameters in this thesis,and two prediction models are constructed to predict porosity and permeability,which are the most important reservoir parameters.Based on this model and combined with actual business requirements,a reservoir parameter prediction system is designed and implemented.Specific research contents include:First,preprocess the logging data and select sensitive features.When selecting features,comprehensively consider the logging data and drilling data,and use the correlation measure method based on Copula function to select the logging curves and drilling parameters with high correlation with porosity and permeability.Then,considering the timing characteristics of logging,drilling data and porosity,a porosity prediction model was constructed based on Sparrow search algorithm optimization bidirectional gated cyclic neural network(SSA-Bi GRU).In order to solve the problem that the parameters of the Bi GRU neural network are complex and it is difficult to find the optimal solution by manual parameter adjustment,the sparrow search algorithm is introduced to optimize the parameters,the Bi GRU neural network model is reconstructed using the optimal parameters,and the porosity is predicted using the data set of Well A in the central depression of Songliao Basin.The experiment shows that the prediction accuracy of SSA-Bi GRU model is 13.5%,5.7% and 4% higher than that of BP,LSTM and Bi GRU models respectively.In view of the low correlation between permeability and logging curves and the more difficult prediction,a bidirectional gated cyclic neural network prediction model integrating one-dimensional convolution neural network is proposed(1DCNN-Bi GRU).The 1DCNN is used to extract the spatial features of logging data,and then the Bi GRU network is used to extract the temporal features to deeply mine the deep features between logging curves and permeability.The experimental results show that the permeability prediction accuracy of1DCNN-Bi GRU model is 8% and 3% higher than that of single CNN model and Bi GRU model respectively.Finally,according to the established porosity and permeability prediction model and the actual reservoir parameter prediction process,the reservoir parameter prediction system is designed and developed for use in the process of oilfield exploration and development.