Research On Production Performance Modelling In Tight Gas Reservoir Based On Data-Driven Method

Tight sand gas is an important part of China’s energy supply.The accurate prediction of production performance of tight gas reservoirs is vital for efficient gas field development.However,the production performances are affected by various complex factors such as geological properties,engineering factors and production restraints in an actual production process.The traditional mechanism-based methods do not work well in forecasting the gas production performance.The data-driven method is able to solve the problems including single means of data processing,high computational cost and complex modelling process by mining the hidden patterns in oil and gas big data.The following technical works were completed:1.Establishing the workflow of production performance modeling of tight gas reservoirs based on literature review.2.Modelling fractured well productivity of tight gas reservoirs.Applying data preprocessing and feature selection to stimulated vertical wells in the eastern Sulige gas field,six variables affecting the absolute open flow potential(AOFP)are determined as follows: flowing bottomhole pressure,perforated thickness,porosity,cumulative flowback fluid volume,matrix permeability and slurry fluid volume.Then,the predictive models of the AOFP were built using a variety of machine learning algorithms.These model performances were evaluated and the best model was determined.3.Interpretation of machine learning-based well productivity model.Three interpretation methods were applied,including SHAP(SHapley Additive ex Planations),partial dependence plot and individual conditional expectation to understand the inferences at both global and local levels,which increase the credibility and practicability of the model.Then,the optimal ranges of the six variables were determined.The particle swarm optimization algorithm was used to optimize the fracturing treatment parameters to improve the gas production of underperformanced wells.4.Modelling production performance of a fractured tight gas reservoir.Based on fractured horizontal wells in the eastern area of Sulige gas field,a tight gas reservoir model was built.Then a reservoir simulator and orthogonal experimental design were applied to simulate the production dynamics under different geological conditions and fracturing treatments.Two predictive models of the tight gas reservoir performance were created using deep neural network algorithms.The gas production rates,reservoir pressure field and gas saturation field were forecasted.