Sensitivity Analysis And Study On Prediction Technology Of Clastic Reservoir In Eastern Kuche

In the process of reservoir sensitivity research,the sensitivity flow evaluation experiment takes a long time.Some experiments cannot make results due to factors such as low core permeability.The experimental data error is large,and the repeatability of parallel experimental data is difficult to guarantee.Due to the limitations of reservoir coring,some evaluation experiments could not be carried out normally.Therefore,for the damage of oil and gas reservoirs,especially the five potential susceptibility of reservoirs,the existing reservoir lithology,physical data and experimental data accumulated in the previous period are used to find out five sensitive damages and reservoirs of the reservoir.The relationship between lithology and physical parameters,explore the inevitable connection between the two,establish a predictive model,normalize and quantify the potential damage factors,and predict the reservoir sensitivity after training the model.And verify the prediction results through experiments.In this paper,through the analysis of reservoir geological structure,the geological factors affecting reservoir sensitivity are used as the independent variables of the sample,and the sensitivity damage index obtained by the sensitivity evaluation experiment is taken as the dependent variable of the sample,and then the appropriate mathematical method is used to establish The prediction model quantitatively predicts five reservoir sensitivity damage indices.This method of prediction eliminates the manpower,material resources,and time spent on sensitivity assessment experiments.The prediction algorithms used in this paper include principal component analysis,polynomial regression,support vector product,random forest and BFGS neural network.Principal component analysis can reduce the data dimension,extract important information of the data,and remove the useless information in the data,thereby increasing the value of the data.Polynomial regression is the most primitive regression method.The calculation process is simple,but the data quality and quantity requirements are high.The support vector machine algorithm shows many unique advantages for processing data with small number,high dimension and obvious nonlinear characteristics.The random forest algorithm is an algorithm that integrates multiple decision trees through the idea of integrated learning.It has extremely high accuracy and can evaluate the importance of each feature.The neural network algorithm has strong nonlinear fitting ability,can map arbitrarily complex nonlinear relationships,and has strong robustness,memory ability and powerful self-learning ability.Based on the selected prediction model and collected samples,a reservoir sensitivity damage index prediction software was developed using C#language,which facilitated the prediction of reservoir sensitivity damage index in the subsequent research process.The model introduces the prediction parameters,which can directly derive the reservoir sensitivity damage type and the damage index,which provides more powerful support for studying the relationship between clay minerals and reservoir sensitivity.The innovation of this paper lies in the choice of predictive model.The introduction of BFGS neural network algorithm,support vector machine algorithm and random forest algorithm improves the accuracy of predictive sensitivity damage index.The BFGS neural network algorithm accelerates the convergence speed of the error function than the BP neural network algorithm.The support vector machine algorithm is suitable for small sample and high dimensional models.The random forest algorithm can solve the multivariate prediction problem well.The influencing factors of reservoir sensitivity There are many corresponding data obtained by geological experiments and sensitive flow evaluation.Therefore,the collected sample data has the characteristics of high dimension and small quantity,so the introduction of the above three algorithms can improve the accuracy of sensitivity damage prediction.Sex.After selecting the above prediction model through the Sklearn database,the C#programming language is used to compile the integration software,which makes the prediction of reservoir sensitivity more convenient and faster.The prediction software saves manpower and material resources consumed by sensitive flow experiments.