Application Research Of Machine Learning In Reservoir Parameter Prediction

Reservoir parameters are important indicators for describing reservoir properties,fluid models and reservoir modeling,and are the basis for detailed evaluation of reservoirs.Traditional reservoir parameter predictions are based on logging curves to establish a model to fit parameters such as porosity and permeability,mainly including regression analysis and empirical formulas.Most of the methods are based on linearity,but in practice.In the environment,the reservoir condition is complex and the heterogeneity is strong.The logging curve and the reservoir parameters are often complex nonlinear relationships.The traditional regression analysis method is used to achieve the expected results of the model is not ideal and the error is better.Large,it is especially necessary to find an efficient and accurate prediction method.In view of the above problems,based on the analysis of three common reservoir parameter prediction models,this paper constructs a support vector machine reservoir parameter prediction model based on particle swarm optimization.The main work includes:Firstly,the nonlinear relationship between the logging curve and the reservoir parameters is analyzed and studied.The support vector machine is suitable for solving the advantages of small sample,nonlinear and high-dimensional problems,and the nonlinear prediction relationship between the two is established.Limit the use and mining of existing logging information.Secondly,the kernel parameter ? and the loss function parameter ?,which have great influence on the support vector machine,are optimized.The particle swarm optimization algorithm has the advantages of fast convergence,high parallelism and strong global search ability,and finds the global optimal support vector.Machine parameters,establish an optimized support vector machine reservoir parameter prediction model.Finally,in order to further verify the practicability and reliability of the model,the optimized and improved model is applied to the prediction of actual reservoir parameters in the study area,and the reservoir porosity and permeability are predicted and evaluated.The experimental results show that the optimized and improved SVM prediction model is superior to traditional regression analysis and neural network in accuracy and running time.The model is stable and effective,and it is reliable and reliable under certain scope and conditions.The potential reservoir parameter prediction method can be used for testing other oil wells in the study area,which will provide some help for the later exploration and development of oil and gas fields.