Development Of Formation Damage Prediction System For Tight Sandstone Reservoirs Based On SVM Method

The tight sandstone reservoirs are characterized by low porosity,low permeability,high capillary pressure,poor pore connectivity and high clay mineral content.During its exploration and development,the possibility,complexity and seriousness of formation damage should not be ignored.Therefore,it is of great significance to predict formation damage quickly and accurately,which provides a basis for formulating or modification the development scheme of oil and gas reservoirs and determination of stimulation measures.The results of formation damage evaluation by physical experiment method are most reliable.However,it is still necessary to seek the formation damage prediction method which is complementary to the physical experiment method because of the difficulty and high cost of the core-flow experiments of tight sandstone reservoirs.Previous practice shows that the prediction of tight sandstone reservoir damage with artificial intelligent methods by using limited experimental data can save a lot of time and cost.The purpose of this paper is to predict the formation damage of tight sandstone reservoirs with small sample data.By studying a variety of artificial intelligent methods,the Support Vector Machine(SVM)is selected as the main method of formation damage prediction,and the internal parameters of the SVM model are optimized by various intelligent methods to improve the prediction accuracy.After these steps,a complete process has been built,and a set of prediction system for tight sandstone formation damage has been developed.Finally,an application instance has been used to detect the system.The following research results were obtained in this paper:(1)It is clear that aqueous trapping damage is the most common and serious damage type in tight sandstone reservoirs;(2)A variety of artificial intelligence methods were selected and the SVM method was used as the main method to predict the damage of tight sandstone reservoirs.It has some special advantages in solving the regression problem of small sample,nonlinear and high dimensional model.(3)A variety of intelligent methods were used to calculate the parameters of the support vector machine prediction model,such as Genetic Algorithm(GA)and Particle Swarm Optimization(PSO),and the prediction accuracy of the SVM was improved by selecting the optimal parameter combination.(4)The damage prediction system of tight sandstone reservoirs was developed by using GUI programming platform of MATLAB.The system had realized 5 main functions,such as user management,data input and processing,reservoir damage prediction,real-time display of operation and user help.The developed system had the advantages of convenient operation and stable operation.(5)The system was used to make a case prediction,which proved that the system had a very good prediction accuracy.The prediction precision of the application Genetic Algorithm was the highest to the three parameters of the penalty factor C,the kernel function parameter ?and the loss factor ? in the SVM model and the effect of application was better.