Research On Stratified Water Injection Volume Prediction Based On Gaussian Process Regression

In oilfield exploitation,in order to ensure the effect of waterflooding oil recovery technology,the layered water injection process is widely used.The determination of layered water injection has a direct impact on the oil production efficiency of the oil field.Therefore,accurate prediction of stratified water injection is extremely important for effectively mobilizing each layer of stratified water injection,improving oil recovery and reducing the inefficient circulation of water between layers.First,this article analyzes the factors that affect the stratified water injection.Collect oilfield productivity data and geological data from oilfields put into production under stable water injection conditions.According to the oil field data distribution and the influence of water injection development,the fluid and reservoir physical property parameters are analyzed.A Gaussian process regression algorithm is used to establish a mathematical model for oilfield stratified water injection prediction.Predict the water injection volume in the oil field and compare it with the actual distribution.Compare the performance difference of different kernel functions in the modeling of oilfield layered water injection prediction model,and determine the kernel function that best fits the law of water injection data.The effectiveness of the stratified water injection prediction model based on Gaussian process regression is verified.Secondly,in order to solve the problems of excessive modeling calculation and too long regression prediction time,the experimental data was analyzed and compared using principal component analysis and nuclear principal component analysis to verify that the nuclear principal component analysis was In terms of aspects,it is possible to retain most of the original information with fewer dimensions.Finally,the kernel function used by the Gaussian process regression algorithm is analyzed,and the influence of hyperparameter changes on the predicted distribution of the model is analyzed.Aiming at the limitation of a single kernel function,the method of combinatorial optimization is used to improve the kernel function,which can effectively enhance the generalization ability of the model and improve the prediction accuracy while ensuring the local learning ability.