Reservoir History Matching For Transient Pressure Based On Support Vector Regression

Well Test Technology in Numerical Reservoir Simulation is that using well bottom pressure infer parameter of stratum and shafts to guide reservoir exploring.History Matching which has being applying in Numerical Reservoir Simulation is important to Well Test Technology.Traditional History Matching has the disadvantages that high cost,bad performance and so on.Support Vector Machine is a new method of machine learning,which has advantages such as strong expansibility,nonlinear and high efficiency.This paper utilizes history matching methodology based on Support Vector Regression to matching and forecasting the parameters in reservoir.The main work in this paper is that building history matching methodology based on support vector regressing,and choosing kernel function depending on ensemble error.The result shows that:1.The inversion methodology of shut data in pressure recovery can be built on the history matching methodology based on support vector regression and the BFGS optimization on reservoir parameter.2.Chooseing kernel function depending on ensemble error can decrease expected risk effectively,avoiding falling into local optimal solution,thus promoting the methodology usability.In Chapter One,we introduces the history of development of Numerical Reservoir Simulation,and History Matching in Numerical Reservoir Simulation.In the end,the current research methods in Numerical Reservoir Simulation and S VM are mentioned.In Chapter Two,we illustrate History Matching in detail,including every segment.In the end,we introduced the method that how to evaluate the result.In Chapter Three,we start with an overview of machine learning,then present the SVM theory including kernel theory.In Chapter Four,we propose the methodology which combines SVM and Well Test Technology,and four experiment are used for proving.Firstly,generating trial examples randomly,then matching for well bottom pressure and its derivative by support vector regression models with different kernel function.Optimizing the model with minimal error in well test analysis to get the optimal parameters.The experiment result will be gotten comparison between the simulated value and the observed value in well bottom pressure and its variation and derivative.The result shows that the methodology can build the model with kernel function for matching well bottom pressure and derivative of pressure.The feature is fast calculation and fitting well.Hence,the methodology worth studying further.