The Identification Of Water-flooded Zone Based On Rough Sets And SVM

At present, most of the water injection development of oil fields have been entered in thelater stages of the moisture in the stratum rising, leading to the complex changes of thereservoir physical and chemical properties, as well as logging curve shape, so there-identification measure of well curves, a new comprehensive evaluation model to accuratelydetermine reservoir flooding level, are of important parts of the scientific guidance to find outthe flooded law and the work of production. In this paper, we’ll combine the attributesreduction in rough set theory with the orderly classification algorithm in support vectormachine theory, aiming to the analyses of the logging data, and prediction the flooding level,which provide a new solution for the flooded layer identification.With continuous development of the geophysical logging methods, well logging data ofthe characterization of reservoir geological information has been greatly enriched. We use theattribute reduction in rough set theory to delete redundant attributes in the data, and to choosesome properties closely associated with the flooding level in the large amounts of data, thenwe can simplify the calculation of the amount of the subsequent classification algorithm whileensuring the accuracy of the results. In this paper, combining with the concept of entropy ininformation theory, we research the discretization of attribute values based on entropy ofinformation. Firstly by understanding the basic concepts of rough set, we use the informationentropy as the evaluation of the importance of the dividing point of the continuous data, anddiscrete sample data as a pretreatment. Then we also use the information entropy as theevaluation of the importance of property, and see the stationary mutual information betweenthe conditions attributes and decision attributes as a principle, then we can reduce theproperties of the discretization result. The experimental results show that the algorithm doesnot reduce the accuracy rate of the orderly classification algorithm of support vector machine, while simplifying the input dimension and training time.Due to the limited data sample, we use support vector machine (SVM) as theclassification method to determine the flooding level. After nearly two decades ofdevelopment, the SVM has been shown to have good classification performance in the case ofsmall samples. SVM, based on the statistical learning theory, with its simple mathematicalform and intuitive geometric interpretation, realizes the principle of structural riskminimization excellently, and has good generalization ability, as well as can effectively avoid"the curse of dimensionality" and local optimum problem. This paper will introduce the basicprinciple of SVM, analyze commonly used classification methods, such as1-v-r,1-v-1, DAG,and KSVM, and contrast their advantages and disadvantages. To aim at the particularity ofclassification of the flooding level, we’ll study in-depth on orderly classification algorithmbased on SVM, and use the embedded space method to build a new hyper plane, whichcontained ordered information, then re-define the form of kernel function, which convert themulti-classification problem to two-classification in the new hyper plane. The SVM, based onembedded space method, makes full use of ordered information of flooding level, instead ofmeaningless symbols, which weld the condition attribute to decision attribute together.At present, whether rough set theory or support vector machine theory has been betterapplied in pattern recognition and in many other areas, but not widely in the flooded layeridentification. This paper will apply discretization and attribute reduction based on rough set,as well as orderly classification algorithm based on SVM embedded space method to theflooded layer identification. By cross-validation of experimental data testing and actualproduction, the results show a high prediction accuracy of the flooding level of the test sample,which proves the proposal in this article has a high recognition rate and better extension.