Prediction Of Reservoir Parameters Based On Attribute Reduction Algorithm Of Rough Sets And SVM

Traditional prediction of reservoir parameters is based on mathematical statistics.forexample, the multiple linear regression,partial least squares regression,well logging theoryand the empirical formula. actual reservoir geological structure in the different presents adifferent regularity in formation lithology, physical properties, electrical properties, oilproperties.log interpretation results should be consistent with their respective laws. it isdifficult to explain the laws by standard of the same set of theoretical or empirical formulas.especially after the decline of the reservoir oil content, oilfield using water injection to extractthe remaining oil, the drive of different layers varying degrees,lead to the four relations of thelayer become more complex.logging curve of the different layers of different flooding levelpresent different changes in the non-linear, ranges, shapes.In view of this situation,support vector machines is asked, this is a machine learningmethods to solve the prediction of reservoir parameters. support vector machine theory isasked since the1990s,and it has been widely recognized and applied.the theory has obviousadvantages in the small sample, nonlinear, high dimensional data fitting or classification. andit overcomes the local extremum problem in the neural network learning algorithm, iteventually converges to the global optimum.Rough set theory is also a data mining tool. it is used in reduction of properties of thedata. rough set theory has obvious advantage in the removal of data noise and deal of the largedata. here we use it to reduce the dimensionality of the data sample.using it to do a datapre-processing for support vector machines. so it makes the max fitting speed.The papers will combine attribute reduction algorithm and support vector machinetogether. the output of the former is the latter’s input. there are many series of the well logscurve actually. different well logs curves have redundancy information, for example, the twocurves have a consistent shape.single logging curve shape does not change obviously.so forthis case,the attribute reduction algorithm will weed out these redundant attributes.so it willguarantee the lost of information never happened and provides streamlined modeling data. soit reduce the time of vector machine learning finally.Using support vector machine fitting method to predict the physical parameters of thereservoir is facing with two problems.the one hand, the choice of kernel function.on the otherhand, the tuning of parameters,after repeated accuracy verification, and finally we decide toadopt the Gaussian radial basis kernel function involved in computing.based on pastexperience,we think that it is the right choice. For adjusting the parameters of the kernel function and the penalty parameter and the sensitive coefficient to regulation, it is importantto follow the principle of the unity of opposites in the law above its qualitative, the usualpractice is to use some learning algorithm to optimize.The ultimate purpose of this paper is to establish the porosity, permeability, saturationinterpretation model. the specific work is to make the physical parameters associate with thewell log curves in the same depth of the well. after fitting the data by support vector machines,predictive model of these three parameters is obtained. in the forecasting process, we need toensure that the prediction accuracy and generalization of the model is uniform and oppositeafter many experiments and reasonable adjustment,we get law between the changes in theparameters and result. and it further be quantified and we get the model finally.After the above theory and practice for the algorithm, we get the model with higherprecision and ir can used widely.it has been used in oil fields for the remaining oil. we get thewide acclaim from customers.